Hugging Face's logo

Spaces:
avans06
/
ImgutilsVideoProcessor
Sleeping

App Files Community
avans06 commited on
Commit
5262ee3
·
1 Parent(s): 494735a

Add application file

Browse files
Files changed (5) hide show
  1. .gitignore +3 -0
  2. README.md +186 -4
  3. app.py +1396 -0
  4. requirements.txt +2 -0
  5. webui.bat +73 -0
.gitignore ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ .vs
2
+ venv
3
+ tmp
README.md CHANGED
@@ -1,14 +1,196 @@
1
  ---
2
- title: VideoProcessorDetection
3
- emoji: 🐠
4
  colorFrom: indigo
5
- colorTo: yellow
6
  sdk: gradio
7
  sdk_version: 5.29.0
8
  app_file: app.py
9
- pinned: false
10
  license: mit
11
  short_description: 'Video Processor: Detection, Classification, Analysis'
12
  ---
13
 
14
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ title: ImgutilsVideoProcessor
3
+ emoji: 🖼️
4
  colorFrom: indigo
5
+ colorTo: pink
6
  sdk: gradio
7
  sdk_version: 5.29.0
8
  app_file: app.py
9
+ pinned: true
10
  license: mit
11
  short_description: 'Video Processor: Detection, Classification, Analysis'
12
  ---
13
 
14
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
15
+
16
+ ---
17
+
18
+ # [deepghs/anime_person_detection](https://huggingface.co/deepghs/anime_person_detection)
19
+
20
+ | Model | FLOPS | Params | F1 Score | Threshold | F1 Plot | Confusion | Labels |
21
+ |:--------------------:|:-------:|:--------:|:----------:|:-----------:|:---------------------------------------------------------------------------------------------------------:|:----------------------------------------------------------------------------------------------------------------------:|:--------:|
22
+ | person_detect_v1.2_s | 3.49k | 11.1M | 0.86 | 0.295 | [plot](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.2_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.2_s/confusion_matrix.png) | `person` |
23
+ | person_detect_v1.3_s | 3.49k | 11.1M | 0.86 | 0.324 | [plot](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.3_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.3_s/confusion_matrix.png) | `person` |
24
+ | person_detect_v0_x | 31.3k | 68.1M | N/A | N/A | N/A | N/A | `person` |
25
+ | person_detect_v0_m | 9.53k | 25.8M | 0.85 | 0.424 | [plot](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v0_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v0_m/confusion_matrix.png) | `person` |
26
+ | person_detect_v0_s | 3.49k | 11.1M | N/A | N/A | N/A | N/A | `person` |
27
+ | person_detect_v1.1_s | 3.49k | 11.1M | 0.86 | 0.384 | [plot](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.1_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.1_s/confusion_matrix.png) | `person` |
28
+ | person_detect_v1.1_n | 898 | 3.01M | 0.85 | 0.327 | [plot](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.1_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.1_n/confusion_matrix.png) | `person` |
29
+ | person_detect_v1.1_m | 9.53k | 25.8M | 0.87 | 0.348 | [plot](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.1_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1.1_m/confusion_matrix.png) | `person` |
30
+ | person_detect_v1_m | 9.53k | 25.8M | 0.86 | 0.351 | [plot](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_person_detection/blob/main/person_detect_v1_m/confusion_matrix.png) | `person` |
31
+
32
+ ___
33
+
34
+ # [deepghs/anime_halfbody_detection](https://huggingface.co/deepghs/anime_halfbody_detection)
35
+
36
+ | Model | FLOPS | Params | F1 Score | Threshold | F1 Plot | Confusion | Labels |
37
+ |:----------------------:|:-------:|:--------:|:----------:|:-----------:|:-------------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------------------------------------------------------:|:----------:|
38
+ | halfbody_detect_v1.0_n | 898 | 3.01M | 0.94 | 0.512 | [plot](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v1.0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v1.0_n/confusion_matrix.png) | `halfbody` |
39
+ | halfbody_detect_v1.0_s | 3.49k | 11.1M | 0.95 | 0.577 | [plot](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v1.0_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v1.0_s/confusion_matrix.png) | `halfbody` |
40
+ | halfbody_detect_v0.4_s | 3.49k | 11.1M | 0.93 | 0.517 | [plot](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v0.4_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v0.4_s/confusion_matrix.png) | `halfbody` |
41
+ | halfbody_detect_v0.3_s | 3.49k | 11.1M | 0.92 | 0.222 | [plot](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v0.3_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v0.3_s/confusion_matrix.png) | `halfbody` |
42
+ | halfbody_detect_v0.2_s | 3.49k | 11.1M | 0.94 | 0.548 | [plot](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v0.2_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_halfbody_detection/blob/main/halfbody_detect_v0.2_s/confusion_matrix.png) | `halfbody` |
43
+
44
+ ___
45
+
46
+ # [deepghs/anime_head_detection](https://huggingface.co/deepghs/anime_head_detection)
47
+
48
+ | Model | Type | FLOPS | Params | F1 Score | Threshold | precision(B) | recall(B) | mAP50(B) | mAP50-95(B) | F1 Plot | Confusion | Labels |
49
+ |:-------------------------:|:------:|:-------:|:--------:|:----------:|:-----------:|:--------------:|:-----------:|:----------:|:-------------:|:------------------------------------------------------------------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------------------------:|:--------:|
50
+ | head_detect_v2.0_x_yv11 | yolo | 195G | 56.9M | 0.93 | 0.458 | 0.95942 | 0.90938 | 0.96853 | 0.78938 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_x_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_x_yv11/confusion_matrix_normalized.png) | `head` |
51
+ | head_detect_v2.0_l_yv11 | yolo | 87.3G | 25.3M | 0.93 | 0.432 | 0.95557 | 0.90905 | 0.9661 | 0.78709 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_l_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_l_yv11/confusion_matrix_normalized.png) | `head` |
52
+ | head_detect_v2.0_m_yv11 | yolo | 68.2G | 20.1M | 0.93 | 0.42 | 0.95383 | 0.90658 | 0.96485 | 0.78511 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_m_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_m_yv11/confusion_matrix_normalized.png) | `head` |
53
+ | head_detect_v2.0_s_yv11 | yolo | 21.5G | 9.43M | 0.92 | 0.383 | 0.954 | 0.89512 | 0.95789 | 0.77753 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_s_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_s_yv11/confusion_matrix_normalized.png) | `head` |
54
+ | head_detect_v2.0_n_yv11 | yolo | 6.44G | 2.59M | 0.91 | 0.365 | 0.94815 | 0.87169 | 0.9452 | 0.75835 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_n_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_n_yv11/confusion_matrix_normalized.png) | `head` |
55
+ | head_detect_v2.0_x | yolo | 227G | 61.6M | 0.93 | 0.459 | 0.95378 | 0.91123 | 0.96593 | 0.78767 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_x/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_x/confusion_matrix_normalized.png) | `head` |
56
+ | head_detect_v2.0_l | yolo | 146G | 39.5M | 0.93 | 0.379 | 0.95124 | 0.91264 | 0.96458 | 0.78627 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_l/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_l/confusion_matrix_normalized.png) | `head` |
57
+ | head_detect_v2.0_m | yolo | 79.1G | 25.9M | 0.93 | 0.397 | 0.95123 | 0.90701 | 0.96403 | 0.78342 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_m/confusion_matrix_normalized.png) | `head` |
58
+ | head_detect_v2.0_s | yolo | 28.6G | 11.1M | 0.92 | 0.413 | 0.95556 | 0.89197 | 0.95799 | 0.77833 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_s/confusion_matrix_normalized.png) | `head` |
59
+ | head_detect_v2.0_n | yolo | 8.19G | 3.01M | 0.91 | 0.368 | 0.94633 | 0.87046 | 0.94361 | 0.75764 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v2.0_n/confusion_matrix_normalized.png) | `head` |
60
+ | head_detect_v1.6_x_rtdetr | rtdetr | 232G | 67.3M | 0.93 | 0.559 | 0.95316 | 0.91697 | 0.96556 | 0.76682 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_x_rtdetr/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_x_rtdetr/confusion_matrix_normalized.png) | `head` |
61
+ | head_detect_v1.6_l_rtdetr | rtdetr | 108G | 32.8M | 0.93 | 0.53 | 0.95113 | 0.90956 | 0.96218 | 0.76201 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_l_rtdetr/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_l_rtdetr/confusion_matrix_normalized.png) | `head` |
62
+ | head_detect_v1.6_s_yv11 | yolo | 21.5G | 9.43M | 0.93 | 0.42 | 0.95273 | 0.90558 | 0.96327 | 0.78566 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s_yv11/confusion_matrix_normalized.png) | `head` |
63
+ | head_detect_v1.6_n_yv11 | yolo | 6.44G | 2.59M | 0.92 | 0.385 | 0.95561 | 0.87798 | 0.95086 | 0.76765 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_n_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_n_yv11/confusion_matrix_normalized.png) | `head` |
64
+ | head_detect_v1.6_s_yv9 | yolo | 22.7G | 6.32M | 0.93 | 0.419 | 0.95464 | 0.90425 | 0.9627 | 0.78663 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s_yv9/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s_yv9/confusion_matrix_normalized.png) | `head` |
65
+ | head_detect_v1.6_t_yv9 | yolo | 6.7G | 1.77M | 0.91 | 0.332 | 0.94968 | 0.8792 | 0.95069 | 0.76789 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_t_yv9/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_t_yv9/confusion_matrix_normalized.png) | `head` |
66
+ | head_detect_v1.6_x | yolo | 258G | 68.2M | 0.94 | 0.448 | 0.9546 | 0.91873 | 0.96878 | 0.79502 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_x/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_x/confusion_matrix_normalized.png) | `head` |
67
+ | head_detect_v1.6_l | yolo | 165G | 43.6M | 0.94 | 0.458 | 0.95733 | 0.92018 | 0.96868 | 0.79428 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_l/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_l/confusion_matrix_normalized.png) | `head` |
68
+ | head_detect_v1.6_s_yv10 | yolo | 24.8G | 8.07M | 0.93 | 0.406 | 0.95424 | 0.90074 | 0.96201 | 0.78713 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s_yv10/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s_yv10/confusion_matrix_normalized.png) | `head` |
69
+ | head_detect_v1.6_n_yv10 | yolo | 8.39G | 2.71M | 0.91 | 0.374 | 0.94845 | 0.87492 | 0.9503 | 0.77059 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_n_yv10/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_n_yv10/confusion_matrix_normalized.png) | `head` |
70
+ | head_detect_v1.6_s | yolo | 28.6G | 11.1M | 0.93 | 0.381 | 0.95333 | 0.90587 | 0.96241 | 0.78688 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_s/confusion_matrix_normalized.png) | `head` |
71
+ | head_detect_v1.6_n | yolo | 8.19G | 3.01M | 0.92 | 0.38 | 0.94835 | 0.88436 | 0.95051 | 0.76766 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.6_n/confusion_matrix_normalized.png) | `head` |
72
+ | head_detect_v1.5_s | yolo | 28.6G | 11.1M | 0.94 | 0.453 | 0.96014 | 0.92275 | 0.96829 | 0.80674 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.5_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.5_s/confusion_matrix_normalized.png) | `head` |
73
+ | head_detect_v1.5_n | yolo | 8.19G | 3.01M | 0.93 | 0.396 | 0.95719 | 0.90511 | 0.9612 | 0.78841 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.5_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.5_n/confusion_matrix_normalized.png) | `head` |
74
+ | head_detect_v1.4_s | yolo | 28.6G | 11.1M | 0.94 | 0.472 | 0.96275 | 0.91875 | 0.96812 | 0.80417 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.4_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.4_s/confusion_matrix_normalized.png) | `head` |
75
+ | head_detect_v1.4_n | yolo | 8.19G | 3.01M | 0.93 | 0.396 | 0.9557 | 0.90559 | 0.96075 | 0.78689 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.4_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.4_n/confusion_matrix_normalized.png) | `head` |
76
+ | head_detect_v1.3_s | yolo | 28.6G | 11.1M | 0.94 | 0.423 | 0.95734 | 0.9257 | 0.97037 | 0.80391 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.3_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.3_s/confusion_matrix_normalized.png) | `head` |
77
+ | head_detect_v1.3_n | yolo | 8.19G | 3.01M | 0.93 | 0.409 | 0.95254 | 0.90674 | 0.96258 | 0.7844 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.3_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.3_n/confusion_matrix_normalized.png) | `head` |
78
+ | head_detect_v1.2_s | yolo | 28.6G | 11.1M | 0.94 | 0.415 | 0.95756 | 0.9271 | 0.97097 | 0.80514 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.2_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.2_s/confusion_matrix_normalized.png) | `head` |
79
+ | head_detect_v1.2_n | yolo | 8.19G | 3.01M | 0.93 | 0.471 | 0.96309 | 0.89766 | 0.9647 | 0.78928 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.2_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.2_n/confusion_matrix_normalized.png) | `head` |
80
+ | head_detect_v1.1_s | yolo | 28.6G | 11.1M | 0.94 | 0.485 | 0.96191 | 0.91892 | 0.97069 | 0.80182 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.1_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.1_s/confusion_matrix_normalized.png) | `head` |
81
+ | head_detect_v1.0_l | yolo | 165G | 43.6M | 0.94 | 0.579 | 0.95881 | 0.91532 | 0.96561 | 0.81417 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_l/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_l/confusion_matrix_normalized.png) | `head` |
82
+ | head_detect_v1.0_x | yolo | 258G | 68.2M | 0.94 | 0.567 | 0.9597 | 0.91947 | 0.96682 | 0.8154 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_x/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_x/confusion_matrix_normalized.png) | `head` |
83
+ | head_detect_v1.0_m | yolo | 79.1G | 25.9M | 0.94 | 0.489 | 0.95805 | 0.9196 | 0.96632 | 0.81383 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_m/confusion_matrix_normalized.png) | `head` |
84
+ | head_detect_v1.0_s | yolo | 28.6G | 11.1M | 0.93 | 0.492 | 0.95267 | 0.91355 | 0.96245 | 0.80371 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_s/confusion_matrix_normalized.png) | `head` |
85
+ | head_detect_v1.0_n | yolo | 8.19G | 3.01M | 0.92 | 0.375 | 0.93999 | 0.9002 | 0.95509 | 0.7849 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v1.0_n/confusion_matrix_normalized.png) | `head` |
86
+ | head_detect_v0.5_s | yolo | 28.6G | 11.1M | 0.92 | 0.415 | 0.93908 | 0.9034 | 0.95697 | 0.77514 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_s/confusion_matrix_normalized.png) | `head` |
87
+ | head_detect_v0.5_n | yolo | 8.19G | 3.01M | 0.91 | 0.446 | 0.93834 | 0.88034 | 0.94784 | 0.75251 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_n/confusion_matrix_normalized.png) | `head` |
88
+ | head_detect_v0.5_s_pruned | yolo | 28.6G | 11.1M | 0.93 | 0.472 | 0.95455 | 0.89865 | 0.9584 | 0.79968 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_s_pruned/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_s_pruned/confusion_matrix_normalized.png) | `head` |
89
+ | head_detect_v0.5_n_pruned | yolo | 8.19G | 3.01M | 0.91 | 0.523 | 0.95254 | 0.8743 | 0.95049 | 0.7807 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_n_pruned/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_n_pruned/confusion_matrix_normalized.png) | `head` |
90
+ | head_detect_v0.5_m_pruned | yolo | 79.1G | 25.9M | 0.94 | 0.52 | 0.9609 | 0.91365 | 0.96501 | 0.81322 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_m_pruned/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.5_m_pruned/confusion_matrix_normalized.png) | `head` |
91
+ | head_detect_v0.4_s | yolo | 28.6G | 11.1M | 0.92 | 0.405 | 0.93314 | 0.90274 | 0.95727 | 0.77193 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.4_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.4_s/confusion_matrix_normalized.png) | `head` |
92
+ | head_detect_v0.4_s_fp | yolo | 28.6G | 11.1M | 0.91 | 0.445 | 0.93181 | 0.89113 | 0.95002 | 0.76302 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.4_s_fp/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.4_s_fp/confusion_matrix_normalized.png) | `head` |
93
+ | head_detect_v0.3_s | yolo | 28.6G | 11.1M | 0.91 | 0.406 | 0.92457 | 0.90351 | 0.95785 | 0.78912 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.3_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.3_s/confusion_matrix_normalized.png) | `head` |
94
+ | head_detect_v0.2_s_plus | yolo | 28.6G | 11.1M | 0.91 | 0.594 | 0.94239 | 0.8774 | 0.94909 | 0.77986 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.2_s_plus/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.2_s_plus/confusion_matrix_normalized.png) | `head` |
95
+ | head_detect_v0.2_s | yolo | 28.6G | 11.1M | 0.9 | 0.461 | 0.91861 | 0.8898 | 0.94765 | 0.77541 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.2_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.2_s/confusion_matrix_normalized.png) | `head` |
96
+ | head_detect_v0.1_s | yolo | 28.6G | 11.1M | 0.9 | 0.504 | 0.91576 | 0.88662 | 0.94213 | 0.7713 | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.1_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0.1_s/confusion_matrix_normalized.png) | `head` |
97
+ | head_detect_v0_n | yolo | 8.19G | 3.01M | 0.9 | 0.316 | N/A | N/A | N/A | N/A | [plot](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_head_detection/blob/main/head_detect_v0_n/confusion_matrix.png) | `head` |
98
+ | head_detect_v0_s | yolo | 28.6G | 11.1M | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | `head` |
99
+
100
+ ___
101
+
102
+ # [deepghs/ccip](https://huggingface.co/deepghs/ccip)
103
+
104
+ | Model | F1 Score | Precision | Recall | Threshold | Cluster_2 | Cluster_Free |
105
+ |:-----------------------------------:|:----------:|:-----------:|:--------:|:-----------:|:-----------:|:--------------:|
106
+ | ccip-caformer_b36-24 | 0.940925 | 0.938254 | 0.943612 | 0.213231 | 0.89508 | 0.957017 |
107
+ | ccip-caformer-24-randaug-pruned | 0.917211 | 0.933481 | 0.901499 | 0.178475 | 0.890366 | 0.922375 |
108
+ | ccip-v2-caformer_s36-10 | 0.906422 | 0.932779 | 0.881513 | 0.207757 | 0.874592 | 0.89241 |
109
+ | ccip-caformer-6-randaug-pruned_fp32 | 0.878403 | 0.893648 | 0.863669 | 0.195122 | 0.810176 | 0.897904 |
110
+ | ccip-caformer-5_fp32 | 0.864363 | 0.90155 | 0.830121 | 0.183973 | 0.792051 | 0.862289 |
111
+ | ccip-caformer-4_fp32 | 0.844967 | 0.870553 | 0.820842 | 0.18367 | 0.795565 | 0.868133 |
112
+ | ccip-caformer_query-12 | 0.823928 | 0.871122 | 0.781585 | 0.141308 | 0.787237 | 0.809426 |
113
+ | ccip-caformer-23_randaug_fp32 | 0.81625 | 0.854134 | 0.781585 | 0.136797 | 0.745697 | 0.8068 |
114
+ | ccip-caformer-2-randaug-pruned_fp32 | 0.78561 | 0.800148 | 0.771592 | 0.171053 | 0.686617 | 0.728195 |
115
+ | ccip-caformer-2_fp32 | 0.755125 | 0.790172 | 0.723055 | 0.141275 | 0.64977 | 0.718516 |
116
+
117
+ * The calculation of `F1 Score`, `Precision`, and `Recall` considers "the characters in both images are the same" as a positive case. `Threshold` is determined by finding the maximum value on the F1 Score curve.
118
+ * `Cluster_2` represents the approximate optimal clustering solution obtained by tuning the eps value in DBSCAN clustering algorithm with min_samples set to `2`, and evaluating the similarity between the obtained clusters and the true distribution using the `random_adjust_score`.
119
+ * `Cluster_Free` represents the approximate optimal solution obtained by tuning the `max_eps` and `min_samples` values in the OPTICS clustering algorithm, and evaluating the similarity between the obtained clusters and the true distribution using the `random_adjust_score`.
120
+
121
+ ___
122
+
123
+ # [deepghs/anime_aesthetic](https://huggingface.co/spaces/deepghs/anime_aesthetic)
124
+
125
+ | Name | FLOPS | Params | Accuracy | AUC | Confusion | Labels |
126
+ |:------------------------:|:-------:|:--------:|:----------:|:------:|:-----------------------------------------------------------------------------------------------------------------:|:----------------------------------------------------------------:|
127
+ | caformer_s36_v0_ls0.2 | 22.10G | 37.22M | 34.68% | 0.7725 | [confusion](https://huggingface.co/deepghs/anime_aesthetic/blob/main/caformer_s36_v0_ls0.2/plot_confusion.png) | `masterpiece`, `best`, `great`, `good`, `normal`, `low`, `worst` |
128
+ | swinv2pv3_v0_448_ls0.2 | 46.20G | 65.94M | 40.32% | 0.8188 | [confusion](https://huggingface.co/deepghs/anime_aesthetic/blob/main/swinv2pv3_v0_448_ls0.2/plot_confusion.png) | `masterpiece`, `best`, `great`, `good`, `normal`, `low`, `worst` |
129
+ | swinv2pv3_v0_448_ls0.2_x | 46.20G | 65.94M | 40.88% | 0.8214 | [confusion](https://huggingface.co/deepghs/anime_aesthetic/blob/main/swinv2pv3_v0_448_ls0.2_x/plot_confusion.png) | `masterpiece`, `best`, `great`, `good`, `normal`, `low`, `worst` |
130
+
131
+ ___
132
+
133
+ # [deepghs/anime_face_detection](https://huggingface.co/deepghs/anime_face_detection)
134
+
135
+ | Model | FLOPS | Params | F1 Score | Threshold | F1 Plot | Confusion | Labels |
136
+ |:------------------:|:-------:|:--------:|:----------:|:-----------:|:-----------------------------------------------------------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------:|:--------:|
137
+ | face_detect_v1.4_n | 898 | 3.01M | 0.94 | 0.278 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.4_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.4_n/confusion_matrix.png) | `face` |
138
+ | face_detect_v1.4_s | 3.49k | 11.1M | 0.95 | 0.307 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.4_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.4_s/confusion_matrix.png) | `face` |
139
+ | face_detect_v1.3_n | 898 | 3.01M | 0.93 | 0.305 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.3_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.3_n/confusion_matrix.png) | `face` |
140
+ | face_detect_v1.2_s | 3.49k | 11.1M | 0.93 | 0.222 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.2_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.2_s/confusion_matrix.png) | `face` |
141
+ | face_detect_v1.3_s | 3.49k | 11.1M | 0.93 | 0.259 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.3_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.3_s/confusion_matrix.png) | `face` |
142
+ | face_detect_v1_s | 3.49k | 11.1M | 0.95 | 0.446 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1_s/confusion_matrix.png) | `face` |
143
+ | face_detect_v1_n | 898 | 3.01M | 0.95 | 0.458 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1_n/confusion_matrix.png) | `face` |
144
+ | face_detect_v0_n | 898 | 3.01M | 0.97 | 0.428 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v0_n/confusion_matrix.png) | `face` |
145
+ | face_detect_v0_s | 3.49k | 11.1M | N/A | N/A | N/A | N/A | `face` |
146
+ | face_detect_v1.1_n | 898 | 3.01M | 0.94 | 0.373 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.1_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.1_n/confusion_matrix.png) | `face` |
147
+ | face_detect_v1.1_s | 3.49k | 11.1M | 0.94 | 0.405 | [plot](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.1_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/anime_face_detection/blob/main/face_detect_v1.1_s/confusion_matrix.png) | `face` |
148
+ ___
149
+
150
+ # [deepghs/real_person_detection](https://huggingface.co/deepghs/real_person_detection)
151
+
152
+ | Model | Type | FLOPS | Params | F1 Score | Threshold | precision(B) | recall(B) | mAP50(B) | mAP50-95(B) | F1 Plot | Confusion | Labels |
153
+ |:-----------------------:|:------:|:-------:|:--------:|:----------:|:-----------:|:--------------:|:-----------:|:----------:|:-------------:|:-----------------------------------------------------------------------------------------------------------:|:-----------------------------------------------------------------------------------------------------------------------------------:|:--------:|
154
+ | person_detect_v0_l_yv11 | yolo | 87.3G | 25.3M | 0.79 | 0.359 | 0.84037 | 0.74055 | 0.82796 | 0.57272 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_l_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_l_yv11/confusion_matrix_normalized.png) | `person` |
155
+ | person_detect_v0_m_yv11 | yolo | 68.2G | 20.1M | 0.78 | 0.351 | 0.83393 | 0.73614 | 0.82195 | 0.56267 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_m_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_m_yv11/confusion_matrix_normalized.png) | `person` |
156
+ | person_detect_v0_s_yv11 | yolo | 21.5G | 9.43M | 0.75 | 0.344 | 0.82356 | 0.6967 | 0.79224 | 0.52304 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_s_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_s_yv11/confusion_matrix_normalized.png) | `person` |
157
+ | person_detect_v0_n_yv11 | yolo | 6.44G | 2.59M | 0.71 | 0.325 | 0.80096 | 0.64148 | 0.74612 | 0.46875 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_n_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_n_yv11/confusion_matrix_normalized.png) | `person` |
158
+ | person_detect_v0_l | yolo | 165G | 43.6M | 0.79 | 0.359 | 0.83674 | 0.74182 | 0.82536 | 0.57022 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_l/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_l/confusion_matrix_normalized.png) | `person` |
159
+ | person_detect_v0_m | yolo | 79.1G | 25.9M | 0.78 | 0.363 | 0.83439 | 0.72529 | 0.81314 | 0.55388 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_m/confusion_matrix_normalized.png) | `person` |
160
+ | person_detect_v0_s | yolo | 28.6G | 11.1M | 0.76 | 0.346 | 0.82522 | 0.69696 | 0.79105 | 0.52201 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_s/confusion_matrix_normalized.png) | `person` |
161
+ | person_detect_v0_n | yolo | 8.19G | 3.01M | 0.72 | 0.32 | 0.80883 | 0.64552 | 0.74996 | 0.47272 | [plot](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_person_detection/blob/main/person_detect_v0_n/confusion_matrix_normalized.png) | `person` |
162
+
163
+ ___
164
+
165
+ # [deepghs/real_head_detection](https://huggingface.co/deepghs/real_head_detection)
166
+
167
+ | Model | Type | FLOPS | Params | F1 Score | Threshold | precision(B) | recall(B) | mAP50(B) | mAP50-95(B) | F1 Plot | Confusion | Labels |
168
+ |:---------------------:|:------:|:-------:|:--------:|:----------:|:-----------:|:--------------:|:-----------:|:----------:|:-------------:|:-------------------------------------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------------------------------------:|:--------:|
169
+ | head_detect_v0_l_yv11 | yolo | 87.3G | 25.3M | 0.81 | 0.199 | 0.90226 | 0.72872 | 0.81049 | 0.5109 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_l_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_l_yv11/confusion_matrix_normalized.png) | `head` |
170
+ | head_detect_v0_m_yv11 | yolo | 68.2G | 20.1M | 0.8 | 0.206 | 0.89855 | 0.72654 | 0.80704 | 0.50804 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_m_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_m_yv11/confusion_matrix_normalized.png) | `head` |
171
+ | head_detect_v0_s_yv11 | yolo | 21.5G | 9.43M | 0.78 | 0.187 | 0.88726 | 0.69234 | 0.77518 | 0.47825 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_s_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_s_yv11/confusion_matrix_normalized.png) | `head` |
172
+ | head_detect_v0_n_yv11 | yolo | 6.44G | 2.59M | 0.74 | 0.14 | 0.87359 | 0.64011 | 0.73393 | 0.44118 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_n_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_n_yv11/confusion_matrix_normalized.png) | `head` |
173
+ | head_detect_v0_l | yolo | 165G | 43.6M | 0.81 | 0.234 | 0.89921 | 0.74092 | 0.81715 | 0.51615 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_l/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_l/confusion_matrix_normalized.png) | `head` |
174
+ | head_detect_v0_m | yolo | 79.1G | 25.9M | 0.8 | 0.228 | 0.90006 | 0.72646 | 0.80614 | 0.50586 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_m/confusion_matrix_normalized.png) | `head` |
175
+ | head_detect_v0_s | yolo | 28.6G | 11.1M | 0.78 | 0.182 | 0.89224 | 0.69382 | 0.77804 | 0.48067 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_s/confusion_matrix_normalized.png) | `head` |
176
+ | head_detect_v0_n | yolo | 8.19G | 3.01M | 0.74 | 0.172 | 0.8728 | 0.64823 | 0.73865 | 0.44501 | [plot](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_head_detection/blob/main/head_detect_v0_n/confusion_matrix_normalized.png) | `head` |
177
+
178
+ ___
179
+
180
+ # [deepghs/real_face_detection](https://huggingface.co/deepghs/real_face_detection)
181
+
182
+ | Model | Type | FLOPS | Params | F1 Score | Threshold | precision(B) | recall(B) | mAP50(B) | mAP50-95(B) | F1 Plot | Confusion | Labels |
183
+ |:---------------------:|:------:|:-------:|:--------:|:----------:|:-----------:|:--------------:|:-----------:|:----------:|:-------------:|:-------------------------------------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------------------------------------:|:--------:|
184
+ | face_detect_v0_s_yv12 | yolo | 21.5G | 9.25M | 0.74 | 0.272 | 0.86931 | 0.6404 | 0.73074 | 0.42652 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_s_yv12/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_s_yv12/confusion_matrix_normalized.png) | `face` |
185
+ | face_detect_v0_n_yv12 | yolo | 6.48G | 2.57M | 0.7 | 0.258 | 0.85246 | 0.59089 | 0.6793 | 0.39182 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_n_yv12/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_n_yv12/confusion_matrix_normalized.png) | `face` |
186
+ | face_detect_v0_l_yv11 | yolo | 87.3G | 25.3M | 0.77 | 0.291 | 0.88458 | 0.67474 | 0.76666 | 0.45722 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_l_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_l_yv11/confusion_matrix_normalized.png) | `face` |
187
+ | face_detect_v0_m_yv11 | yolo | 68.2G | 20.1M | 0.76 | 0.262 | 0.87947 | 0.67315 | 0.76073 | 0.45288 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_m_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_m_yv11/confusion_matrix_normalized.png) | `face` |
188
+ | face_detect_v0_s_yv11 | yolo | 21.5G | 9.43M | 0.73 | 0.271 | 0.87001 | 0.63572 | 0.72683 | 0.42706 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_s_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_s_yv11/confusion_matrix_normalized.png) | `face` |
189
+ | face_detect_v0_n_yv11 | yolo | 6.44G | 2.59M | 0.7 | 0.263 | 0.86044 | 0.58577 | 0.67641 | 0.38975 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_n_yv11/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_n_yv11/confusion_matrix_normalized.png) | `face` |
190
+ | face_detect_v0_l | yolo | 165G | 43.6M | 0.76 | 0.277 | 0.87894 | 0.67335 | 0.76313 | 0.4532 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_l/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_l/confusion_matrix_normalized.png) | `face` |
191
+ | face_detect_v0_m | yolo | 79.1G | 25.9M | 0.75 | 0.277 | 0.87687 | 0.66265 | 0.75114 | 0.44262 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_m/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_m/confusion_matrix_normalized.png) | `face` |
192
+ | face_detect_v0_s | yolo | 28.6G | 11.1M | 0.73 | 0.282 | 0.86932 | 0.63557 | 0.72494 | 0.42219 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_s/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_s/confusion_matrix_normalized.png) | `face` |
193
+ | face_detect_v0_n | yolo | 8.19G | 3.01M | 0.7 | 0.257 | 0.85337 | 0.58877 | 0.67471 | 0.38692 | [plot](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_n/F1_curve.png) | [confusion](https://huggingface.co/deepghs/real_face_detection/blob/main/face_detect_v0_n/confusion_matrix_normalized.png) | `face` |
194
+
195
+ ___
196
+
app.py ADDED
@@ -0,0 +1,1396 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import io
2
+ import os
3
+ import gc
4
+ import re
5
+ import cv2
6
+ import time
7
+ import zipfile
8
+ import tempfile
9
+ import traceback
10
+ import numpy as np
11
+ import gradio as gr
12
+ import imgutils.detect.person as person_detector
13
+ import imgutils.detect.halfbody as halfbody_detector
14
+ import imgutils.detect.head as head_detector
15
+ import imgutils.detect.face as face_detector
16
+ import imgutils.metrics.ccip as ccip_analyzer
17
+ import imgutils.metrics.dbaesthetic as dbaesthetic_analyzer
18
+ import imgutils.metrics.lpips as lpips_module
19
+ from PIL import Image
20
+ from typing import List, Tuple, Dict, Any, Union, Optional, Iterator
21
+
22
+ # --- Constants for File Types ---
23
+ IMAGE_EXTENSIONS = ('.png', '.jpg', '.jpeg', '.webp', '.bmp', '.tiff', '.tif', '.gif')
24
+ VIDEO_EXTENSIONS = ('.mp4', '.avi', '.mov', '.mkv', '.flv', '.webm', '.mpeg', '.mpg')
25
+
26
+ # --- Helper Functions ---
27
+ def sanitize_filename(filename: str, max_len: int = 50) -> str:
28
+ """Removes invalid characters and shortens a filename for safe use."""
29
+ # Remove path components
30
+ base_name = os.path.basename(filename)
31
+ # Remove extension
32
+ name_part, _ = os.path.splitext(base_name)
33
+ # Replace spaces and problematic characters with underscores
34
+ sanitized = re.sub(r'[\\/*?:"<>|\s]+', '_', name_part)
35
+ # Remove leading/trailing underscores/periods
36
+ sanitized = sanitized.strip('._')
37
+ # Limit length (important for temp paths and OS limits)
38
+ sanitized = sanitized[:max_len]
39
+ # Ensure it's not empty after sanitization
40
+ if not sanitized:
41
+ return "file"
42
+ return sanitized
43
+
44
+ def convert_to_pil(frame: np.ndarray) -> Image.Image:
45
+ """Converts an OpenCV frame (BGR) to a PIL Image (RGB)."""
46
+ # Add error handling for potentially empty frames
47
+ if frame is None or frame.size == 0:
48
+ raise ValueError("Cannot convert empty frame to PIL Image")
49
+ try:
50
+ return Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
51
+ except Exception as e:
52
+ # Re-raise with more context if conversion fails
53
+ raise RuntimeError(f"Failed to convert frame to PIL Image: {e}")
54
+
55
+ def image_to_bytes(img: Image.Image, format: str = 'PNG') -> bytes:
56
+ """Converts a PIL Image to bytes."""
57
+ if img is None:
58
+ raise ValueError("Cannot convert None image to bytes")
59
+ byte_arr = io.BytesIO()
60
+ img.save(byte_arr, format=format)
61
+ return byte_arr.getvalue()
62
+
63
+ def create_zip_file(image_data: Dict[str, bytes], output_path: str) -> None:
64
+ """
65
+ Creates a zip file containing the provided images directly at the output_path.
66
+
67
+ Args:
68
+ image_data: A dictionary where keys are filenames (including paths within zip)
69
+ and values are image bytes.
70
+ output_path: The full path where the zip file should be created.
71
+ """
72
+ if not image_data:
73
+ raise ValueError("No image data provided to create zip file.")
74
+ if not output_path:
75
+ raise ValueError("No output path provided for the zip file.")
76
+
77
+ print(f"Creating zip file at: {output_path}")
78
+
79
+ try:
80
+ # Ensure parent directory exists (useful if output_path is nested)
81
+ # Though NamedTemporaryFile usually handles this for its own path.
82
+ parent_dir = os.path.dirname(output_path)
83
+ if parent_dir: # Check if there is a parent directory component
84
+ os.makedirs(parent_dir, exist_ok=True)
85
+
86
+ with zipfile.ZipFile(output_path, 'w', zipfile.ZIP_DEFLATED) as zipf:
87
+ # Sort items for potentially better organization and predictability
88
+ for filename, img_bytes in sorted(image_data.items()):
89
+ zipf.writestr(filename, img_bytes)
90
+ print(f"Successfully created zip file with {len(image_data)} items at {output_path}.")
91
+ # No return value needed as we are writing to a path
92
+ except Exception as e:
93
+ print(f"Error creating zip file at {output_path}: {e}")
94
+ # If zip creation fails, attempt to remove the partially created file
95
+ if os.path.exists(output_path):
96
+ try:
97
+ os.remove(output_path)
98
+ print(f"Removed partially created/failed zip file: {output_path}")
99
+ except OSError as remove_err:
100
+ print(f"Warning: Could not remove failed zip file {output_path}: {remove_err}")
101
+ raise # Re-raise the original exception
102
+
103
+ def generate_filename(
104
+ base_name: str, # Should be the core identifier, e.g., "frame_X_person_Y_scoreZ"
105
+ aesthetic_label: Optional[str] = None,
106
+ ccip_cluster_id_for_lpips_logic: Optional[int] = None, # Original CCIP ID, used to decide if LPIPS is sub-cluster
107
+ ccip_folder_naming_index: Optional[int] = None, # The new 000, 001, ... index based on image count
108
+ source_prefix_for_ccip_folder: Optional[str] = None, # The source filename prefix for CCIP folder
109
+ lpips_folder_naming_index: Optional[Union[int, str]] = None, # New: Can be int (0,1,2...) or "noise"
110
+ file_extension: str = '.png',
111
+ # Suffix flags for this specific image:
112
+ is_halfbody_primary_target_type: bool = False, # If this image itself was a halfbody primary target
113
+ is_derived_head_crop: bool = False,
114
+ is_derived_face_crop: bool = False,
115
+ ) -> str:
116
+ """
117
+ Generates the final filename, incorporating aesthetic label, cluster directory,
118
+ and crop indicators. CCIP and LPIPS folder names are sorted by image count.
119
+ """
120
+ filename_stem = base_name
121
+
122
+ # Add suffixes for derived crops.
123
+ # For halfbody primary targets, the base_name should already contain "halfbody".
124
+ # This flag is more for potentially adding a suffix if desired, but currently not used to add a suffix.
125
+ # if is_halfbody_primary_target_type:
126
+ # filename_stem += "_halfbody" # Potentially redundant if base_name good.
127
+
128
+ if is_derived_head_crop:
129
+ filename_stem += "_headCrop"
130
+ if is_derived_face_crop:
131
+ filename_stem += "_faceCrop"
132
+
133
+ filename_with_extension = filename_stem + file_extension
134
+
135
+ path_parts = []
136
+ # New CCIP folder naming based on source prefix and sorted index
137
+ if ccip_folder_naming_index is not None and source_prefix_for_ccip_folder is not None:
138
+ path_parts.append(f"{source_prefix_for_ccip_folder}_ccip_{ccip_folder_naming_index:03d}")
139
+
140
+ # LPIPS folder naming based on the new sorted index or "noise"
141
+ if lpips_folder_naming_index is not None:
142
+ lpips_folder_name_part_str: Optional[str] = None
143
+ if isinstance(lpips_folder_naming_index, str) and lpips_folder_naming_index == "noise":
144
+ lpips_folder_name_part_str = "noise"
145
+ elif isinstance(lpips_folder_naming_index, int):
146
+ lpips_folder_name_part_str = f"{lpips_folder_naming_index:03d}"
147
+
148
+ if lpips_folder_name_part_str is not None:
149
+ # Determine prefix based on whether the item was originally in a CCIP cluster
150
+ if ccip_cluster_id_for_lpips_logic is not None: # LPIPS is sub-cluster if item had an original CCIP ID
151
+ lpips_folder_name_base = "lpips_sub_"
152
+ else: # No CCIP, LPIPS is primary
153
+ lpips_folder_name_base = "lpips_"
154
+ path_parts.append(f"{lpips_folder_name_base}{lpips_folder_name_part_str}")
155
+
156
+ final_filename_part = filename_with_extension
157
+ if aesthetic_label:
158
+ final_filename_part = f"{aesthetic_label}_{filename_with_extension}"
159
+
160
+ if path_parts:
161
+ return f"{'/'.join(path_parts)}/{final_filename_part}"
162
+ else:
163
+ return final_filename_part
164
+
165
+ # --- Core Processing Function for a single source (video or image sequence) ---
166
+ def _process_input_source_frames(
167
+ source_file_prefix: str, # Sanitized name for this source (e.g., "myvideo" or "ImageGroup123")
168
+ # Iterator yielding: (PIL.Image, frame_identifier_string, current_item_index, total_items_for_desc)
169
+ # For videos, current_item_index is the 1-based raw frame number.
170
+ # For images, current_item_index is the 1-based image number in the sequence.
171
+ frames_provider: Iterator[Tuple[Image.Image, int, int, int]],
172
+ is_video_source: bool, # To adjust some logging/stats messages
173
+ # Person Detection
174
+ enable_person_detection: bool,
175
+ min_target_width_person_percentage: float,
176
+ person_model_name: str,
177
+ person_conf_threshold: float,
178
+ person_iou_threshold: float,
179
+ # Half-Body Detection
180
+ enable_halfbody_detection: bool,
181
+ enable_halfbody_cropping: bool,
182
+ min_target_width_halfbody_percentage: float,
183
+ halfbody_model_name: str,
184
+ halfbody_conf_threshold: float,
185
+ halfbody_iou_threshold: float,
186
+ # Head Detection
187
+ enable_head_detection: bool,
188
+ enable_head_cropping: bool,
189
+ min_crop_width_head_percentage: float,
190
+ enable_head_filtering: bool,
191
+ head_model_name: str,
192
+ head_conf_threshold: float,
193
+ head_iou_threshold: float,
194
+ # Face Detection
195
+ enable_face_detection: bool,
196
+ enable_face_cropping: bool,
197
+ min_crop_width_face_percentage: float,
198
+ enable_face_filtering: bool,
199
+ face_model_name: str,
200
+ face_conf_threshold: float,
201
+ face_iou_threshold: float,
202
+ # CCIP Classification
203
+ enable_ccip_classification: bool,
204
+ ccip_model_name: str,
205
+ ccip_threshold: float,
206
+ # LPIPS Clustering
207
+ enable_lpips_clustering: bool,
208
+ lpips_threshold: float,
209
+ # Aesthetic Analysis
210
+ enable_aesthetic_analysis: bool,
211
+ aesthetic_model_name: str,
212
+ # Gradio Progress (specific to this source's processing)
213
+ progress_updater # Function: (progress_value: float, desc: str) -> None
214
+ ) -> Tuple[str | None, str]:
215
+ """
216
+ Processes frames from a given source (video or image sequence) according to the specified parameters.
217
+ Order: Person => Half-Body (alternative) => Face Detection => Head Detection => CCIP => Aesthetic.
218
+
219
+ Returns:
220
+ A tuple containing:
221
+ - Path to the output zip file (or None if error).
222
+ - Status message string.
223
+ """
224
+ # This list will hold data for images that pass all filters, BEFORE LPIPS and final zipping
225
+ images_pending_final_processing: List[Dict[str, Any]] = []
226
+
227
+ # CCIP specific data
228
+ ccip_clusters_info: List[Tuple[int, np.ndarray]] = []
229
+ next_ccip_cluster_id = 0
230
+
231
+ # Stats
232
+ processed_items_count = 0
233
+ total_persons_detected_raw, total_halfbodies_detected_raw = 0, 0
234
+ person_targets_processed_count, halfbody_targets_processed_count, fullframe_targets_processed_count = 0, 0, 0
235
+ total_faces_detected_on_targets, total_heads_detected_on_targets = 0, 0
236
+
237
+ # These count items added to images_pending_final_processing
238
+ main_targets_pending_count, face_crops_pending_count, head_crops_pending_count = 0, 0, 0
239
+ items_filtered_by_face_count, items_filtered_by_head_count = 0, 0
240
+ ccip_applied_count, aesthetic_applied_count = 0, 0
241
+ # LPIPS stats
242
+ lpips_images_subject_to_clustering, total_lpips_clusters_created, total_lpips_noise_samples = 0, 0, 0
243
+
244
+ gc_interval = 100 # items from provider
245
+ start_time = time.time()
246
+
247
+ # Progress update for initializing this specific video
248
+ progress_updater(0, desc=f"Initializing {source_file_prefix}...")
249
+ output_zip_path_temp = None
250
+ output_zip_path_final = None
251
+
252
+ try:
253
+ # --- Main Loop for processing items from the frames_provider ---
254
+ for pil_image_full_frame, frame_specific_index, current_item_index, total_items_for_desc in frames_provider:
255
+ progress_value_for_updater = (current_item_index) / total_items_for_desc if total_items_for_desc > 0 else 1.0
256
+
257
+
258
+ # The description string should reflect what current_item_index means
259
+ item_description = ""
260
+ if is_video_source:
261
+ # For video, total_items_in_source_for_description is total raw frames.
262
+ # current_item_index is the raw frame index of the *sampled* frame.
263
+ # We also need a counter for *sampled* frames for a "processed X of Y (sampled)" message.
264
+ # processed_items_count counts sampled frames.
265
+ item_description = f"Scanning frame {current_item_index}/{total_items_for_desc} (processed {processed_items_count + 1} sampled)"
266
+
267
+ else: # For images
268
+ item_description = f"image {current_item_index}/{total_items_for_desc}"
269
+
270
+ progress_updater(
271
+ min(progress_value_for_updater, 1.0), # Cap progress at 1.0
272
+ desc=f"Processing {item_description} for {source_file_prefix}"
273
+ )
274
+ # processed_items_count still counts how many items are yielded by the provider
275
+ # (i.e., how many sampled frames for video, or how many images for image sequence)
276
+ processed_items_count += 1
277
+
278
+ try:
279
+ full_frame_width = pil_image_full_frame.width # Store for percentage calculations
280
+ print(f"--- Processing item ID {frame_specific_index} (Width: {full_frame_width}px) for {source_file_prefix} ---")
281
+
282
+ # List to hold PIL images that are the primary subjects for this frame
283
+ # Each element: {'pil': Image, 'base_name': str, 'source_type': 'person'/'halfbody'/'fullframe'}
284
+ primary_targets_for_frame: List[Dict[str, Any]] = []
285
+ processed_primary_source_this_frame = False # Flag if Person or HalfBody yielded targets
286
+
287
+ # --- 1. Person Detection ---
288
+ if enable_person_detection and full_frame_width > 0:
289
+ print(" Attempting Person Detection...")
290
+ min_person_target_px_width = full_frame_width * min_target_width_person_percentage
291
+ person_detections = person_detector.detect_person(
292
+ pil_image_full_frame, model_name=person_model_name,
293
+ conf_threshold=person_conf_threshold, iou_threshold=person_iou_threshold
294
+ )
295
+ total_persons_detected_raw += len(person_detections)
296
+ if person_detections:
297
+ print(f" Detected {len(person_detections)} raw persons.")
298
+ valid_person_targets = 0
299
+ for i, (bbox, _, score) in enumerate(person_detections):
300
+ # Check width before full crop for minor optimization
301
+ detected_person_width = bbox[2] - bbox[0]
302
+ if detected_person_width >= min_person_target_px_width:
303
+ primary_targets_for_frame.append({
304
+ 'pil': pil_image_full_frame.crop(bbox),
305
+ 'base_name': f"{source_file_prefix}_item_{frame_specific_index}_person_{i}_score{int(score*100)}",
306
+ 'source_type': 'person'})
307
+ person_targets_processed_count +=1
308
+ valid_person_targets +=1
309
+ else:
310
+ print(f" Person {i} width {detected_person_width}px < min {min_person_target_px_width:.0f}px. Skipping.")
311
+ if valid_person_targets > 0:
312
+ processed_primary_source_this_frame = True
313
+ print(f" Added {valid_person_targets} persons as primary targets.")
314
+
315
+ # --- 2. Half-Body Detection (if Person not processed and HBD enabled) ---
316
+ if not processed_primary_source_this_frame and enable_halfbody_detection and full_frame_width > 0:
317
+ print(" Attempting Half-Body Detection (on full item)...")
318
+ min_halfbody_target_px_width = full_frame_width * min_target_width_halfbody_percentage
319
+ halfbody_detections = halfbody_detector.detect_halfbody(
320
+ pil_image_full_frame, model_name=halfbody_model_name,
321
+ conf_threshold=halfbody_conf_threshold, iou_threshold=halfbody_iou_threshold
322
+ )
323
+ total_halfbodies_detected_raw += len(halfbody_detections)
324
+ if halfbody_detections:
325
+ print(f" Detected {len(halfbody_detections)} raw half-bodies.")
326
+ valid_halfbody_targets = 0
327
+ for i, (bbox, _, score) in enumerate(halfbody_detections):
328
+ detected_hb_width = bbox[2] - bbox[0]
329
+ # Cropping must be enabled and width must be sufficient for it to be a target
330
+ if enable_halfbody_cropping and detected_hb_width >= min_halfbody_target_px_width:
331
+ primary_targets_for_frame.append({
332
+ 'pil': pil_image_full_frame.crop(bbox),
333
+ 'base_name': f"{source_file_prefix}_item_{frame_specific_index}_halfbody_{i}_score{int(score*100)}",
334
+ 'source_type': 'halfbody'})
335
+ halfbody_targets_processed_count +=1
336
+ valid_halfbody_targets +=1
337
+ elif enable_halfbody_cropping:
338
+ print(f" Half-body {i} width {detected_hb_width}px < min {min_halfbody_target_px_width:.0f}px. Skipping.")
339
+ if valid_halfbody_targets > 0:
340
+ processed_primary_source_this_frame = True
341
+ print(f" Added {valid_halfbody_targets} half-bodies as primary targets.")
342
+
343
+ # --- 3. Full Frame/Image (fallback) ---
344
+ if not processed_primary_source_this_frame:
345
+ print(" Processing Full Item as primary target.")
346
+ primary_targets_for_frame.append({
347
+ 'pil': pil_image_full_frame.copy(),
348
+ 'base_name': f"{source_file_prefix}_item_{frame_specific_index}_full",
349
+ 'source_type': 'fullframe'})
350
+ fullframe_targets_processed_count += 1
351
+
352
+ # --- Process each identified primary_target_for_frame ---
353
+ for target_data in primary_targets_for_frame:
354
+ current_pil: Image.Image = target_data['pil']
355
+ current_base_name: str = target_data['base_name'] # Base name for this main target
356
+ current_source_type: str = target_data['source_type']
357
+ current_pil_width = current_pil.width # For sub-crop percentage calculations
358
+ print(f" Processing target: {current_base_name} (type: {current_source_type}, width: {current_pil_width}px)")
359
+
360
+ # Store PILs of successful crops from current_pil for this target
361
+ keep_this_target = True
362
+ item_area = current_pil_width * current_pil.height
363
+ potential_face_crops_pil: List[Image.Image] = []
364
+ potential_head_crops_pil: List[Image.Image] = []
365
+
366
+ # --- A. Face Detection ---
367
+ if keep_this_target and enable_face_detection and current_pil_width > 0:
368
+ print(f" Detecting faces in {current_base_name}...")
369
+ min_face_crop_px_width = current_pil_width * min_crop_width_face_percentage
370
+ face_detections = face_detector.detect_faces(
371
+ current_pil, model_name=face_model_name,
372
+ conf_threshold=face_conf_threshold, iou_threshold=face_iou_threshold
373
+ )
374
+ total_faces_detected_on_targets += len(face_detections)
375
+ if not face_detections and enable_face_filtering:
376
+ keep_this_target = False
377
+ items_filtered_by_face_count += 1
378
+ print(f" FILTERING TARGET {current_base_name} (no face).")
379
+ elif face_detections and enable_face_cropping:
380
+ for f_idx, (f_bbox, _, _) in enumerate(face_detections):
381
+ if (f_bbox[2]-f_bbox[0]) >= min_face_crop_px_width:
382
+ potential_face_crops_pil.append(current_pil.crop(f_bbox))
383
+ else:
384
+ print(f" Face {f_idx} too small. Skipping crop.")
385
+
386
+ # --- B. Head Detection ---
387
+ if keep_this_target and enable_head_detection and current_pil_width > 0:
388
+ print(f" Detecting heads in {current_base_name}...")
389
+ min_head_crop_px_width = current_pil_width * min_crop_width_head_percentage
390
+ head_detections = head_detector.detect_heads(
391
+ current_pil, model_name=head_model_name,
392
+ conf_threshold=head_conf_threshold, iou_threshold=head_iou_threshold
393
+ )
394
+ total_heads_detected_on_targets += len(head_detections)
395
+ if not head_detections and enable_head_filtering:
396
+ keep_this_target = False
397
+ items_filtered_by_head_count += 1
398
+ print(f" FILTERING TARGET {current_base_name} (no head).")
399
+ potential_face_crops_pil.clear() # Clear faces if head filter removed target
400
+ elif head_detections and enable_head_cropping:
401
+ for h_idx, (h_bbox, _, _) in enumerate(head_detections):
402
+ h_w = h_bbox[2]-h_bbox[0] # h_h = h_bbox[3]-h_bbox[1]
403
+ if h_w >= min_head_crop_px_width and item_area > 0:
404
+ potential_head_crops_pil.append(current_pil.crop(h_bbox))
405
+ else:
406
+ print(f" Head {h_idx} too small or too large relative to parent. Skipping crop.")
407
+
408
+ # --- If target is filtered, clean up and skip to next target ---
409
+ if not keep_this_target:
410
+ print(f" Target {current_base_name} was filtered by face/head presence rules. Discarding it and its potential crops.")
411
+ if current_pil is not None:
412
+ del current_pil
413
+ potential_face_crops_pil.clear()
414
+ potential_head_crops_pil.clear()
415
+ continue # To the next primary_target_for_frame
416
+
417
+ # --- C. CCIP Classification (on current_pil, if it's kept) ---
418
+ assigned_ccip_id = None # This is the original CCIP ID
419
+ if enable_ccip_classification:
420
+ print(f" Classifying {current_base_name} with CCIP...")
421
+ try:
422
+ feature = ccip_analyzer.ccip_extract_feature(current_pil, model=ccip_model_name)
423
+ best_match_cid = None
424
+ min_diff = float('inf')
425
+ # Find the best potential match among existing clusters
426
+ if ccip_clusters_info: # Only loop if there are clusters to compare against
427
+ for cid, rep_f in ccip_clusters_info:
428
+ diff = ccip_analyzer.ccip_difference(feature, rep_f, model=ccip_model_name)
429
+ if diff < min_diff:
430
+ min_diff = diff
431
+ best_match_cid = cid
432
+
433
+ # Decide whether to use the best match or create a new cluster
434
+ if best_match_cid is not None and min_diff < ccip_threshold:
435
+ assigned_ccip_id = best_match_cid
436
+ print(f" -> Matched Cluster {assigned_ccip_id} (Diff: {min_diff:.6f} <= Threshold {ccip_threshold:.3f})")
437
+ else:
438
+ # No suitable match found (either no clusters existed, or the best match's diff was strictly greater than threshold)
439
+ # Create a new cluster
440
+ assigned_ccip_id = next_ccip_cluster_id
441
+ ccip_clusters_info.append((assigned_ccip_id, feature))
442
+ if not ccip_clusters_info or len(ccip_clusters_info) == 1:
443
+ print(f" -> New Cluster {assigned_ccip_id} (First item or no prior suitable clusters)")
444
+ else:
445
+ # MODIFIED: Log message reflecting that new cluster is formed if diff > threshold
446
+ print(f" -> New Cluster {assigned_ccip_id} (Min diff to others: {min_diff:.6f} > Threshold {ccip_threshold:.3f})")
447
+ next_ccip_cluster_id += 1
448
+ print(f" CCIP: Target {current_base_name} -> Original Cluster ID {assigned_ccip_id}")
449
+ del feature
450
+ ccip_applied_count += 1
451
+ except Exception as e_ccip:
452
+ print(f" Error CCIP: {e_ccip}")
453
+
454
+ # --- D. Aesthetic Analysis (on current_pil, if it's kept) ---
455
+ item_aesthetic_label = None
456
+ if enable_aesthetic_analysis:
457
+ print(f" Analyzing {current_base_name} for aesthetics...")
458
+ try:
459
+ res = dbaesthetic_analyzer.anime_dbaesthetic(current_pil, model_name=aesthetic_model_name)
460
+ if isinstance(res, tuple) and len(res) >= 1:
461
+ item_aesthetic_label = res[0]
462
+ print(f" Aesthetic: Target {current_base_name} -> {item_aesthetic_label}")
463
+ aesthetic_applied_count += 1
464
+ except Exception as e_aes:
465
+ print(f" Error Aesthetic: {e_aes}")
466
+
467
+ add_current_pil_to_pending_list = True
468
+ if current_source_type == 'fullframe':
469
+ can_skip_fullframe_target = False
470
+ if enable_face_detection or enable_head_detection:
471
+ found_valid_sub_crop_from_enabled_detector = False
472
+ if enable_face_detection and len(potential_face_crops_pil) > 0:
473
+ found_valid_sub_crop_from_enabled_detector = True
474
+
475
+ if not found_valid_sub_crop_from_enabled_detector and \
476
+ enable_head_detection and len(potential_head_crops_pil) > 0:
477
+ found_valid_sub_crop_from_enabled_detector = True
478
+
479
+ if not found_valid_sub_crop_from_enabled_detector: # No valid crops from any enabled sub-detector
480
+ can_skip_fullframe_target = True # All enabled sub-detectors failed
481
+
482
+ if can_skip_fullframe_target:
483
+ add_current_pil_to_pending_list = False
484
+ print(f" Skipping save of fullframe target '{current_base_name}' because all enabled sub-detectors (Face/Head) yielded no valid-width crops.")
485
+
486
+ if add_current_pil_to_pending_list:
487
+ # --- E. Save current_pil (if it passed all filters) ---
488
+ # Add main target to pending list
489
+ images_pending_final_processing.append({
490
+ 'pil_image': current_pil.copy(), 'base_name_for_filename': current_base_name,
491
+ 'ccip_cluster_id': assigned_ccip_id, 'aesthetic_label': item_aesthetic_label,
492
+ 'is_halfbody_primary_target_type': (current_source_type == 'halfbody'),
493
+ 'is_derived_head_crop': False, 'is_derived_face_crop': False,
494
+ 'lpips_cluster_id': None, # Will be filled by LPIPS clustering
495
+ 'lpips_folder_naming_index': None # Will be filled by LPIPS renaming
496
+ })
497
+ main_targets_pending_count +=1
498
+
499
+ # --- F. Save Face Crops (derived from current_pil) ---
500
+ for i, fc_pil in enumerate(potential_face_crops_pil):
501
+ images_pending_final_processing.append({
502
+ 'pil_image': fc_pil, 'base_name_for_filename': f"{current_base_name}_face{i}",
503
+ 'ccip_cluster_id': assigned_ccip_id, 'aesthetic_label': item_aesthetic_label,
504
+ 'is_halfbody_primary_target_type': False,
505
+ 'is_derived_head_crop': False, 'is_derived_face_crop': True,
506
+ 'lpips_cluster_id': None,
507
+ 'lpips_folder_naming_index': None
508
+ })
509
+ face_crops_pending_count+=1
510
+ potential_face_crops_pil.clear()
511
+
512
+ # --- G. Save Head Crops (derived from current_pil) ---
513
+ for i, hc_pil in enumerate(potential_head_crops_pil):
514
+ images_pending_final_processing.append({
515
+ 'pil_image': hc_pil, 'base_name_for_filename': f"{current_base_name}_head{i}",
516
+ 'ccip_cluster_id': assigned_ccip_id, 'aesthetic_label': item_aesthetic_label,
517
+ 'is_halfbody_primary_target_type': False,
518
+ 'is_derived_head_crop': True, 'is_derived_face_crop': False,
519
+ 'lpips_cluster_id': None,
520
+ 'lpips_folder_naming_index': None
521
+ })
522
+ head_crops_pending_count+=1
523
+ potential_head_crops_pil.clear()
524
+
525
+ if current_pil is not None: # Ensure current_pil exists before attempting to delete
526
+ del current_pil # Clean up the PIL for this target_data
527
+
528
+ primary_targets_for_frame.clear()
529
+ except Exception as item_proc_err:
530
+ print(f"!! Major Error processing item ID {frame_specific_index} for {source_file_prefix}: {item_proc_err}")
531
+ traceback.print_exc()
532
+ # Cleanup local vars for this item if error
533
+ if 'primary_targets_for_frame' in locals():
534
+ primary_targets_for_frame.clear()
535
+ # Also ensure current_pil from inner loop is cleaned up if error happened mid-loop
536
+ if 'current_pil' in locals() and current_pil is not None:
537
+ del current_pil
538
+
539
+ if processed_items_count % gc_interval == 0:
540
+ gc.collect()
541
+ print(f" [GC triggered at {processed_items_count} items for {source_file_prefix}]")
542
+ # --- End of Main Item Processing Loop ---
543
+
544
+ print(f"\nRunning final GC before LPIPS/Zipping for {source_file_prefix}...")
545
+ gc.collect()
546
+
547
+ if not images_pending_final_processing:
548
+ status_message = f"Processing for {source_file_prefix} finished, but no images were generated or passed filters for LPIPS/Zipping."
549
+ print(status_message)
550
+ return None, status_message
551
+
552
+ # --- LPIPS Clustering Stage ---
553
+ print(f"\n--- LPIPS Clustering Stage for {source_file_prefix} (Images pending: {len(images_pending_final_processing)}) ---")
554
+ if enable_lpips_clustering:
555
+ print(f" LPIPS Clustering enabled with threshold: {lpips_threshold}")
556
+ lpips_images_subject_to_clustering = len(images_pending_final_processing)
557
+
558
+ if enable_ccip_classification and next_ccip_cluster_id > 0: # CCIP was used
559
+ print(" LPIPS clustering within CCIP clusters.")
560
+ images_by_ccip: Dict[Optional[int], List[int]] = {} # ccip_id -> list of original indices
561
+ for i, item_data in enumerate(images_pending_final_processing):
562
+ ccip_id = item_data['ccip_cluster_id'] # Original CCIP ID
563
+ if ccip_id not in images_by_ccip:
564
+ images_by_ccip[ccip_id] = []
565
+ images_by_ccip[ccip_id].append(i)
566
+
567
+ for ccip_id, indices_in_ccip_cluster in images_by_ccip.items():
568
+ pils_for_lpips_sub_cluster = [images_pending_final_processing[idx]['pil_image'] for idx in indices_in_ccip_cluster]
569
+ if len(pils_for_lpips_sub_cluster) > 1:
570
+ print(f" Clustering {len(pils_for_lpips_sub_cluster)} images in CCIP cluster {ccip_id}...")
571
+ try:
572
+ lpips_sub_ids = lpips_module.lpips_clustering(pils_for_lpips_sub_cluster, threshold=lpips_threshold)
573
+ for i_sub, lpips_id in enumerate(lpips_sub_ids):
574
+ original_idx = indices_in_ccip_cluster[i_sub]
575
+ images_pending_final_processing[original_idx]['lpips_cluster_id'] = lpips_id
576
+ except Exception as e_lpips_sub:
577
+ print(f" Error LPIPS sub-cluster CCIP {ccip_id}: {e_lpips_sub}")
578
+ elif len(pils_for_lpips_sub_cluster) == 1:
579
+ images_pending_final_processing[indices_in_ccip_cluster[0]]['lpips_cluster_id'] = 0 # type: ignore
580
+ del images_by_ccip
581
+ if 'pils_for_lpips_sub_cluster' in locals():
582
+ del pils_for_lpips_sub_cluster # Ensure cleanup
583
+ else: # LPIPS on all images globally
584
+ print(" LPIPS clustering on all collected images.")
585
+ all_pils_for_global_lpips = [item['pil_image'] for item in images_pending_final_processing]
586
+ if len(all_pils_for_global_lpips) > 1:
587
+ try:
588
+ lpips_global_ids = lpips_module.lpips_clustering(all_pils_for_global_lpips, threshold=lpips_threshold)
589
+ for i, lpips_id in enumerate(lpips_global_ids):
590
+ images_pending_final_processing[i]['lpips_cluster_id'] = lpips_id
591
+ except Exception as e_lpips_global:
592
+ print(f" Error LPIPS global: {e_lpips_global}")
593
+ elif len(all_pils_for_global_lpips) == 1:
594
+ images_pending_final_processing[0]['lpips_cluster_id'] = 0 # type: ignore
595
+ del all_pils_for_global_lpips
596
+
597
+ # Calculate LPIPS stats
598
+ all_final_lpips_ids = [item.get('lpips_cluster_id') for item in images_pending_final_processing if item.get('lpips_cluster_id') is not None]
599
+ if all_final_lpips_ids:
600
+ unique_lpips_clusters = set(filter(lambda x: x != -1, all_final_lpips_ids))
601
+ total_lpips_clusters_created = len(unique_lpips_clusters)
602
+ total_lpips_noise_samples = sum(1 for x in all_final_lpips_ids if x == -1)
603
+ else:
604
+ print(" LPIPS Clustering disabled.")
605
+
606
+ # --- CCIP Folder Renaming Logic ---
607
+ original_ccip_id_to_new_naming_index: Dict[int, int] = {}
608
+ if enable_ccip_classification:
609
+ print(f" Preparing CCIP folder renaming for {source_file_prefix}...")
610
+ ccip_image_counts: Dict[int, int] = {} # original_ccip_id -> count of images in it
611
+ for item_data_for_count in images_pending_final_processing:
612
+ original_ccip_id_val = item_data_for_count.get('ccip_cluster_id')
613
+ if original_ccip_id_val is not None:
614
+ ccip_image_counts[original_ccip_id_val] = ccip_image_counts.get(original_ccip_id_val, 0) + 1
615
+
616
+ if ccip_image_counts:
617
+ # Sort original ccip_ids by their counts in descending order
618
+ sorted_ccip_groups_by_count: List[Tuple[int, int]] = sorted(
619
+ ccip_image_counts.items(),
620
+ key=lambda item: item[1], # Sort by count
621
+ reverse=True
622
+ )
623
+ for new_idx, (original_id, count) in enumerate(sorted_ccip_groups_by_count):
624
+ original_ccip_id_to_new_naming_index[original_id] = new_idx
625
+ print(f" CCIP Remap for {source_file_prefix}: Original ID {original_id} (count: {count}) -> New Naming Index {new_idx:03d}")
626
+ else:
627
+ print(f" No CCIP-assigned images found for {source_file_prefix} to perform renaming.")
628
+
629
+ # --- LPIPS Folder Renaming Logic ---
630
+ if enable_lpips_clustering:
631
+ print(f" Preparing LPIPS folder renaming for {source_file_prefix}...")
632
+ # Initialize/Reset lpips_folder_naming_index for all items
633
+ for item_data in images_pending_final_processing:
634
+ item_data['lpips_folder_naming_index'] = None
635
+
636
+ if enable_ccip_classification and next_ccip_cluster_id > 0: # LPIPS within CCIP
637
+ print(f" LPIPS renaming within CCIP clusters for {source_file_prefix}.")
638
+ items_grouped_by_original_ccip: Dict[Optional[int], List[Dict[str, Any]]] = {}
639
+ for item_data in images_pending_final_processing:
640
+ original_ccip_id = item_data.get('ccip_cluster_id')
641
+ if original_ccip_id not in items_grouped_by_original_ccip: items_grouped_by_original_ccip[original_ccip_id] = []
642
+ items_grouped_by_original_ccip[original_ccip_id].append(item_data)
643
+
644
+ for original_ccip_id, items_in_ccip in items_grouped_by_original_ccip.items():
645
+ lpips_counts_in_ccip: Dict[int, int] = {} # original_lpips_id (non-noise) -> count
646
+ for item_data in items_in_ccip:
647
+ lpips_id = item_data.get('lpips_cluster_id')
648
+ if lpips_id is not None and lpips_id != -1:
649
+ lpips_counts_in_ccip[lpips_id] = lpips_counts_in_ccip.get(lpips_id, 0) + 1
650
+
651
+ lpips_id_to_naming_in_ccip: Dict[int, Union[int, str]] = {}
652
+ if lpips_counts_in_ccip:
653
+ sorted_lpips = sorted(lpips_counts_in_ccip.items(), key=lambda x: x[1], reverse=True)
654
+ for new_idx, (lpips_id, count) in enumerate(sorted_lpips):
655
+ lpips_id_to_naming_in_ccip[lpips_id] = new_idx
656
+ ccip_disp = f"OrigCCIP-{original_ccip_id}" if original_ccip_id is not None else "NoCCIP"
657
+ print(f" LPIPS Remap in {ccip_disp}: OrigLPIPS ID {lpips_id} (count: {count}) -> New Naming Index {new_idx:03d}")
658
+
659
+ for item_data in items_in_ccip:
660
+ lpips_id = item_data.get('lpips_cluster_id')
661
+ if lpips_id is not None:
662
+ if lpips_id == -1: item_data['lpips_folder_naming_index'] = "noise"
663
+ elif lpips_id in lpips_id_to_naming_in_ccip:
664
+ item_data['lpips_folder_naming_index'] = lpips_id_to_naming_in_ccip[lpips_id]
665
+ del items_grouped_by_original_ccip
666
+ else: # Global LPIPS
667
+ print(f" Global LPIPS renaming for {source_file_prefix}.")
668
+ global_lpips_counts: Dict[int, int] = {}
669
+ for item_data in images_pending_final_processing:
670
+ lpips_id = item_data.get('lpips_cluster_id')
671
+ if lpips_id is not None and lpips_id != -1:
672
+ global_lpips_counts[lpips_id] = global_lpips_counts.get(lpips_id, 0) + 1
673
+
674
+ global_lpips_id_to_naming: Dict[int, Union[int, str]] = {}
675
+ if global_lpips_counts:
676
+ sorted_global_lpips = sorted(global_lpips_counts.items(), key=lambda x: x[1], reverse=True)
677
+ for new_idx, (lpips_id, count) in enumerate(sorted_global_lpips):
678
+ global_lpips_id_to_naming[lpips_id] = new_idx
679
+ print(f" Global LPIPS Remap: OrigLPIPS ID {lpips_id} (count: {count}) -> New Naming Index {new_idx:03d}")
680
+
681
+ for item_data in images_pending_final_processing:
682
+ lpips_id = item_data.get('lpips_cluster_id')
683
+ if lpips_id is not None:
684
+ if lpips_id == -1: item_data['lpips_folder_naming_index'] = "noise"
685
+ elif lpips_id in global_lpips_id_to_naming:
686
+ item_data['lpips_folder_naming_index'] = global_lpips_id_to_naming[lpips_id]
687
+ gc.collect()
688
+
689
+ # --- Final Zipping Stage ---
690
+ images_to_zip: Dict[str, bytes] = {}
691
+ print(f"\n--- Final Zipping Stage for {source_file_prefix} ({len(images_pending_final_processing)} items) ---")
692
+ for item_data in images_pending_final_processing:
693
+ original_ccip_id_for_item = item_data.get('ccip_cluster_id')
694
+ current_ccip_naming_idx_for_folder: Optional[int] = None
695
+
696
+ if enable_ccip_classification and original_ccip_id_for_item is not None and \
697
+ original_ccip_id_for_item in original_ccip_id_to_new_naming_index:
698
+ current_ccip_naming_idx_for_folder = original_ccip_id_to_new_naming_index[original_ccip_id_for_item]
699
+
700
+ current_lpips_naming_idx_for_folder = item_data.get('lpips_folder_naming_index')
701
+
702
+ final_filename = generate_filename(
703
+ base_name=item_data['base_name_for_filename'],
704
+ aesthetic_label=item_data.get('aesthetic_label'),
705
+ ccip_cluster_id_for_lpips_logic=original_ccip_id_for_item,
706
+ ccip_folder_naming_index=current_ccip_naming_idx_for_folder,
707
+ source_prefix_for_ccip_folder=source_file_prefix if current_ccip_naming_idx_for_folder is not None else None,
708
+ lpips_folder_naming_index=current_lpips_naming_idx_for_folder,
709
+ is_halfbody_primary_target_type=item_data['is_halfbody_primary_target_type'],
710
+ is_derived_head_crop=item_data['is_derived_head_crop'],
711
+ is_derived_face_crop=item_data['is_derived_face_crop']
712
+ )
713
+ try:
714
+ images_to_zip[final_filename] = image_to_bytes(item_data['pil_image'])
715
+ except Exception as e_bytes:
716
+ print(f" Error converting/adding {final_filename} to zip: {e_bytes}")
717
+ finally:
718
+ if 'pil_image' in item_data and item_data['pil_image'] is not None:
719
+ del item_data['pil_image']
720
+ images_pending_final_processing.clear()
721
+
722
+ if not images_to_zip:
723
+ status_message = f"Processing for {source_file_prefix} finished, but no images were converted for zipping."
724
+ print(status_message)
725
+ return None, status_message
726
+
727
+ print(f"Preparing zip file for {source_file_prefix} with {len(images_to_zip)} images...")
728
+ progress_updater(1.0, desc=f"Creating Zip File for {source_file_prefix}...")
729
+ zip_start_time = time.time()
730
+
731
+ # Use NamedTemporaryFile with delete=False for the final output path
732
+ # This file will persist until manually cleaned or OS cleanup
733
+ temp_zip_file = tempfile.NamedTemporaryFile(delete=False, suffix=".zip")
734
+ output_zip_path_temp = temp_zip_file.name
735
+ temp_zip_file.close() # Close the handle, but file remains
736
+
737
+ try:
738
+ # Write data to the temporary file path
739
+ create_zip_file(images_to_zip, output_zip_path_temp)
740
+ zip_duration = time.time() - zip_start_time
741
+ print(f"Temporary zip file for {source_file_prefix} created in {zip_duration:.2f} seconds at {output_zip_path_temp}")
742
+
743
+ # Construct the new, desired filename
744
+ temp_dir = os.path.dirname(output_zip_path_temp)
745
+ timestamp = int(time.time())
746
+ desired_filename = f"{source_file_prefix}_processed_{timestamp}.zip"
747
+ output_zip_path_final = os.path.join(temp_dir, desired_filename)
748
+
749
+ # Rename the temporary file to the desired name
750
+ print(f"Renaming temp file for {source_file_prefix} to: {output_zip_path_final}")
751
+ os.rename(output_zip_path_temp, output_zip_path_final)
752
+ print("Rename successful.")
753
+ output_zip_path_temp = None # Clear temp path as it's been renamed
754
+
755
+ except Exception as zip_or_rename_err:
756
+ print(f"Error during zip creation or renaming for {source_file_prefix}: {zip_or_rename_err}")
757
+ # Clean up the *original* temp file if it still exists and renaming failed
758
+ if output_zip_path_temp and os.path.exists(output_zip_path_temp):
759
+ try:
760
+ os.remove(output_zip_path_temp)
761
+ except OSError:
762
+ pass
763
+ if output_zip_path_final and os.path.exists(output_zip_path_final): # Check if rename partially happened
764
+ try:
765
+ os.remove(output_zip_path_final)
766
+ except OSError:
767
+ pass
768
+ raise zip_or_rename_err # Re-raise the error
769
+
770
+ # --- Prepare Status Message ---
771
+ processing_duration = time.time() - start_time - zip_duration # Exclude zipping time from processing time
772
+ total_duration = time.time() - start_time # Includes zipping/renaming
773
+
774
+ # --- Build final status message ---
775
+ person_stats = "N/A"
776
+ if enable_person_detection:
777
+ person_stats = f"{total_persons_detected_raw} raw, {person_targets_processed_count} targets (>{min_target_width_person_percentage*100:.1f}% itemW)"
778
+
779
+ halfbody_stats = "N/A"
780
+ if enable_halfbody_detection:
781
+ halfbody_stats = f"{total_halfbodies_detected_raw} raw, {halfbody_targets_processed_count} targets (>{min_target_width_halfbody_percentage*100:.1f}% itemW)"
782
+ fullframe_stats = f"{fullframe_targets_processed_count} targets"
783
+
784
+ face_stats = "N/A"
785
+ if enable_face_detection:
786
+ face_stats = f"{total_faces_detected_on_targets} on targets, {face_crops_pending_count} crops pending (>{min_crop_width_face_percentage*100:.1f}% parentW)"
787
+ if enable_face_filtering:
788
+ face_stats += f", {items_filtered_by_face_count} targets filtered"
789
+
790
+ head_stats = "N/A"
791
+ if enable_head_detection:
792
+ head_stats = f"{total_heads_detected_on_targets} on targets, {head_crops_pending_count} crops pending (>{min_crop_width_head_percentage*100:.1f}% parentW)"
793
+ if enable_head_filtering:
794
+ head_stats += f", {items_filtered_by_head_count} targets filtered"
795
+
796
+ ccip_stats = "N/A"
797
+ if enable_ccip_classification:
798
+ ccip_stats = f"{next_ccip_cluster_id} original clusters created, on {ccip_applied_count} targets. Folders renamed by image count."
799
+
800
+ lpips_stats = "N/A"
801
+ if enable_lpips_clustering:
802
+ lpips_stats = f"{lpips_images_subject_to_clustering} images processed, {total_lpips_clusters_created} clusters, {total_lpips_noise_samples} noise. Folders renamed by image count."
803
+
804
+ aesthetic_stats = "N/A"
805
+ if enable_aesthetic_analysis:
806
+ aesthetic_stats = f"On {aesthetic_applied_count} targets"
807
+
808
+ item_desc_for_stats = "Items from Provider" if not is_video_source else "Sampled Frames"
809
+ status_message = (
810
+ f"Processing for '{source_file_prefix}' Complete!\n"
811
+ f"Total time: {total_duration:.2f}s (Proc: {processing_duration:.2f}s, Zip: {zip_duration:.2f}s)\n"
812
+ f"{item_desc_for_stats}: {total_items_for_desc}, Processed Items: {processed_items_count}\n"
813
+ f"--- Primary Targets Processed ---\n"
814
+ f" Person Detection: {person_stats}\n"
815
+ f" Half-Body Detection: {halfbody_stats}\n"
816
+ f" Full Item Processing: {fullframe_stats}\n"
817
+ f"--- Items Pending Final Processing ({main_targets_pending_count} main, {face_crops_pending_count} face, {head_crops_pending_count} head) ---\n"
818
+ f" Face Detection: {face_stats}\n"
819
+ f" Head Detection: {head_stats}\n"
820
+ f" CCIP Classification: {ccip_stats}\n"
821
+ f" LPIPS Clustering: {lpips_stats}\n"
822
+ f" Aesthetic Analysis: {aesthetic_stats}\n"
823
+ f"Zip file contains {len(images_to_zip)} images.\n"
824
+ f"Output Zip: {output_zip_path_final}"
825
+ )
826
+ print(status_message)
827
+ progress_updater(1.0, desc=f"Finished {source_file_prefix}!")
828
+
829
+ # Return the path to the zip file
830
+ return output_zip_path_final, status_message
831
+
832
+ except Exception as e:
833
+ print(f"!! An unhandled error occurred during processing of {source_file_prefix}: {e}")
834
+ traceback.print_exc() # Print detailed traceback for debugging
835
+ # Clean up main data structures
836
+ images_pending_final_processing.clear()
837
+ ccip_clusters_info.clear()
838
+ gc.collect()
839
+
840
+ # Clean up temp file if it exists on general error
841
+ if output_zip_path_temp and os.path.exists(output_zip_path_temp):
842
+ try:
843
+ os.remove(output_zip_path_temp)
844
+ except OSError:
845
+ pass
846
+
847
+ # Clean up final file if it exists on general error (maybe renaming succeeded but later code failed)
848
+ if output_zip_path_final and os.path.exists(output_zip_path_final):
849
+ try:
850
+ os.remove(output_zip_path_final)
851
+ except OSError:
852
+ pass
853
+ return None, f"An error occurred with {source_file_prefix}: {e}"
854
+
855
+ # --- Main Processing Function for Input files ---
856
+ def process_inputs_main(
857
+ input_file_objects: List[Any], # Gradio File component gives list of tempfile._TemporaryFileWrapper
858
+ sample_interval_ms: int, # Relevant for videos only
859
+ # Person Detection
860
+ enable_person_detection: bool,
861
+ min_target_width_person_percentage: float,
862
+ person_model_name: str,
863
+ person_conf_threshold: float,
864
+ person_iou_threshold: float,
865
+ # Half-Body Detection
866
+ enable_halfbody_detection: bool,
867
+ enable_halfbody_cropping: bool,
868
+ min_target_width_halfbody_percentage: float,
869
+ halfbody_model_name: str,
870
+ halfbody_conf_threshold: float,
871
+ halfbody_iou_threshold: float,
872
+ # Head Detection
873
+ enable_head_detection: bool,
874
+ enable_head_cropping: bool,
875
+ min_crop_width_head_percentage: float,
876
+ enable_head_filtering: bool,
877
+ head_model_name: str,
878
+ head_conf_threshold: float,
879
+ head_iou_threshold: float,
880
+ # Face Detection
881
+ enable_face_detection: bool,
882
+ enable_face_cropping: bool,
883
+ min_crop_width_face_percentage: float,
884
+ enable_face_filtering: bool,
885
+ face_model_name: str,
886
+ face_conf_threshold: float,
887
+ face_iou_threshold: float,
888
+ # CCIP Classification
889
+ enable_ccip_classification: bool,
890
+ ccip_model_name: str,
891
+ ccip_threshold: float,
892
+ # LPIPS Clustering
893
+ enable_lpips_clustering: bool,
894
+ lpips_threshold: float,
895
+ # Aesthetic Analysis
896
+ enable_aesthetic_analysis: bool,
897
+ aesthetic_model_name: str,
898
+ progress=gr.Progress(track_tqdm=True) # Gradio progress for overall processing
899
+ ) -> Tuple[Optional[List[str]], str]: # Returns list of ZIP paths and combined status
900
+
901
+ if not input_file_objects:
902
+ return [], "Error: No files provided."
903
+
904
+ video_file_temp_objects: List[Any] = []
905
+ image_file_temp_objects: List[Any] = []
906
+
907
+ for file_obj in input_file_objects:
908
+ # gr.Files returns a list of tempfile._TemporaryFileWrapper objects
909
+ # We need the .name attribute to get the actual file path
910
+ file_name = getattr(file_obj, 'orig_name', file_obj.name) # Use original name if available
911
+ if isinstance(file_name, str):
912
+ lower_file_name = file_name.lower()
913
+ if any(lower_file_name.endswith(ext) for ext in VIDEO_EXTENSIONS):
914
+ video_file_temp_objects.append(file_obj)
915
+ elif any(lower_file_name.endswith(ext) for ext in IMAGE_EXTENSIONS):
916
+ image_file_temp_objects.append(file_obj)
917
+ else:
918
+ print(f"Warning: File '{file_name}' has an unrecognized extension and will be skipped.")
919
+ else:
920
+ print(f"Warning: File object {file_obj} does not have a valid name and will be skipped.")
921
+
922
+
923
+ output_zip_paths_all_sources = []
924
+ all_status_messages = []
925
+
926
+ total_processing_tasks = (1 if image_file_temp_objects else 0) + len(video_file_temp_objects)
927
+ if total_processing_tasks == 0:
928
+ return [], "No processable video or image files found in the input."
929
+
930
+ tasks_completed_count = 0
931
+
932
+ # Print overall settings once
933
+ print(f"--- Overall Batch Processing Settings ---")
934
+ print(f" Number of image sequences to process: {1 if image_file_temp_objects else 0}")
935
+ print(f" Number of videos to process: {len(video_file_temp_objects)}")
936
+ print(f" Sample Interval (for videos): {sample_interval_ms}ms")
937
+ print(f" Detection Order: Person => Half-Body (alt) => Face => Head. Then: CCIP => LPIPS => Aesthetic.")
938
+ print(f" Person Detect = {enable_person_detection}" + (f" (MinW:{min_target_width_person_percentage*100:.1f}%, Mdl:{person_model_name}, Conf:{person_conf_threshold:.2f}, IoU:{person_iou_threshold:.2f})" if enable_person_detection else ""))
939
+ print(f" HalfBody Detect = {enable_halfbody_detection}" + (f" (FullFrameOnly, Crop:{enable_halfbody_cropping}, MinW:{min_target_width_halfbody_percentage*100:.1f}%, Mdl:{halfbody_model_name}, Conf:{halfbody_conf_threshold:.2f}, IoU:{halfbody_iou_threshold:.2f})" if enable_halfbody_detection else ""))
940
+ print(f" Face Detect = {enable_face_detection}" + (f" (Crop:{enable_face_cropping}, MinW:{min_crop_width_face_percentage*100:.1f}%, Filter:{enable_face_filtering}, Mdl:{face_model_name}, Conf:{face_conf_threshold:.2f}, IoU:{face_iou_threshold:.2f})" if enable_face_detection else ""))
941
+ print(f" Head Detect = {enable_head_detection}" + (f" (Crop:{enable_head_cropping}, MinW:{min_crop_width_head_percentage*100:.1f}%, Filter:{enable_head_filtering}, Mdl:{head_model_name}, Conf:{head_conf_threshold:.2f}, IoU:{head_iou_threshold:.2f})" if enable_head_detection else ""))
942
+ print(f" CCIP Classify = {enable_ccip_classification}" + (f" (Mdl:{ccip_model_name}, Thr:{ccip_threshold:.3f})" if enable_ccip_classification else ""))
943
+ print(f" LPIPS Clustering = {enable_lpips_clustering}" + (f" (Thr:{lpips_threshold:.3f})" if enable_lpips_clustering else ""))
944
+ print(f" Aesthetic Analyze = {enable_aesthetic_analysis}" + (f" (Mdl:{aesthetic_model_name})" if enable_aesthetic_analysis else ""))
945
+ print(f"--- End of Overall Settings ---")
946
+
947
+
948
+ # --- Process Image Sequence (if any) ---
949
+ if image_file_temp_objects:
950
+ image_group_label_base = "ImageGroup"
951
+ # Attempt to use first image name for more uniqueness, fallback to timestamp
952
+ try:
953
+ first_image_orig_name = getattr(image_file_temp_objects[0], 'orig_name', image_file_temp_objects[0].name)
954
+ image_group_label_base = sanitize_filename(first_image_orig_name, max_len=20)
955
+ except:
956
+ pass # Stick with "ImageGroup"
957
+
958
+ image_source_file_prefix = f"{image_group_label_base}_{int(time.time())}"
959
+
960
+ current_task_number = tasks_completed_count + 1
961
+ progress_description_prefix = f"Image Seq. {current_task_number}/{total_processing_tasks} ({image_source_file_prefix})"
962
+ progress(tasks_completed_count / total_processing_tasks, desc=f"{progress_description_prefix}: Starting...")
963
+ print(f"\n>>> Processing Image Sequence: {image_source_file_prefix} ({len(image_file_temp_objects)} images) <<<")
964
+
965
+ def image_frames_provider_generator() -> Iterator[Tuple[Image.Image, int, int, int]]:
966
+ num_images = len(image_file_temp_objects)
967
+ for idx, img_obj in enumerate(image_file_temp_objects):
968
+ try:
969
+ pil_img = Image.open(img_obj.name).convert('RGB')
970
+ yield pil_img, idx, idx + 1, num_images
971
+ except Exception as e_load:
972
+ print(f"Error loading image {getattr(img_obj, 'orig_name', img_obj.name)}: {e_load}. Skipping.")
973
+ # If we skip, the total_items_in_source for _process_input_source_frames might be off
974
+ # For simplicity, we'll proceed, but this could be refined to adjust total_items dynamically.
975
+ # Or, pre-filter loadable images. For now, just skip.
976
+ continue
977
+
978
+ def image_group_progress_updater(item_progress_value: float, desc: str):
979
+ overall_progress = (tasks_completed_count + item_progress_value) / total_processing_tasks
980
+ progress(overall_progress, desc=f"{progress_description_prefix}: {desc}")
981
+
982
+ try:
983
+ zip_file_path_single, status_message_single = _process_input_source_frames(
984
+ source_file_prefix=image_source_file_prefix,
985
+ frames_provider=image_frames_provider_generator(),
986
+ is_video_source=False,
987
+ enable_person_detection=enable_person_detection,
988
+ min_target_width_person_percentage=min_target_width_person_percentage,
989
+ person_model_name=person_model_name,
990
+ person_conf_threshold=person_conf_threshold,
991
+ person_iou_threshold=person_iou_threshold,
992
+ enable_halfbody_detection=enable_halfbody_detection,
993
+ enable_halfbody_cropping=enable_halfbody_cropping,
994
+ min_target_width_halfbody_percentage=min_target_width_halfbody_percentage,
995
+ halfbody_model_name=halfbody_model_name,
996
+ halfbody_conf_threshold=halfbody_conf_threshold,
997
+ halfbody_iou_threshold=halfbody_iou_threshold,
998
+ enable_head_detection=enable_head_detection,
999
+ enable_head_cropping=enable_head_cropping,
1000
+ min_crop_width_head_percentage=min_crop_width_head_percentage,
1001
+ enable_head_filtering=enable_head_filtering,
1002
+ head_model_name=head_model_name,
1003
+ head_conf_threshold=head_conf_threshold,
1004
+ head_iou_threshold=head_iou_threshold,
1005
+ enable_face_detection=enable_face_detection,
1006
+ enable_face_cropping=enable_face_cropping,
1007
+ min_crop_width_face_percentage=min_crop_width_face_percentage,
1008
+ enable_face_filtering=enable_face_filtering,
1009
+ face_model_name=face_model_name,
1010
+ face_conf_threshold=face_conf_threshold,
1011
+ face_iou_threshold=face_iou_threshold,
1012
+ enable_ccip_classification=enable_ccip_classification,
1013
+ ccip_model_name=ccip_model_name,
1014
+ ccip_threshold=ccip_threshold,
1015
+ enable_lpips_clustering=enable_lpips_clustering,
1016
+ lpips_threshold=lpips_threshold,
1017
+ enable_aesthetic_analysis=enable_aesthetic_analysis,
1018
+ aesthetic_model_name=aesthetic_model_name,
1019
+ progress_updater=image_group_progress_updater
1020
+ )
1021
+ if zip_file_path_single:
1022
+ output_zip_paths_all_sources.append(zip_file_path_single)
1023
+ all_status_messages.append(f"--- Image Sequence ({image_source_file_prefix}) Processing Succeeded ---\n{status_message_single}")
1024
+ else:
1025
+ all_status_messages.append(f"--- Image Sequence ({image_source_file_prefix}) Processing Failed ---\n{status_message_single}")
1026
+ except Exception as e_img_seq:
1027
+ error_msg = f"Critical error during processing of image sequence {image_source_file_prefix}: {e_img_seq}"
1028
+ print(error_msg)
1029
+ traceback.print_exc()
1030
+ all_status_messages.append(f"--- Image Sequence ({image_source_file_prefix}) Processing CRITICALLY FAILED ---\n{error_msg}")
1031
+
1032
+ tasks_completed_count += 1
1033
+ print(f">>> Finished attempt for Image Sequence: {image_source_file_prefix} <<<")
1034
+
1035
+ # --- Process Video Files (if any) ---
1036
+ for video_idx, video_file_temp_obj in enumerate(video_file_temp_objects):
1037
+ video_path_temp = video_file_temp_obj.name
1038
+ video_original_filename = os.path.basename(getattr(video_file_temp_obj, 'orig_name', video_path_temp))
1039
+ video_source_file_prefix = sanitize_filename(video_original_filename)
1040
+
1041
+ current_task_number = tasks_completed_count + 1
1042
+ progress_description_prefix = f"Video {current_task_number}/{total_processing_tasks}"
1043
+
1044
+ print(f"\n>>> Processing Video: {video_original_filename} (Sanitized Prefix: {video_source_file_prefix}) <<<")
1045
+ progress(tasks_completed_count / total_processing_tasks, desc=f"{progress_description_prefix}: Starting processing...")
1046
+
1047
+ # It yields: (PIL.Image, frame_identifier_string, current_raw_frame_index_from_video, total_items_for_desc)
1048
+ # The third element will be the raw frame number based on CAP_PROP_POS_FRAMES or current_pos_ms
1049
+ # to align progress with total_items_for_desc (raw frame count).
1050
+ def video_frames_provider_generator(video_path: str, interval_ms: int) -> Iterator[Tuple[Image.Image, int, int, int]]:
1051
+ cap = cv2.VideoCapture(video_path)
1052
+ if not cap.isOpened():
1053
+ print(f"Error: Could not open video file for provider: {video_path}")
1054
+ return
1055
+
1056
+ total_items_for_desc = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
1057
+
1058
+ if total_items_for_desc <= 0:
1059
+ print(f"Warning: Video {video_original_filename} reported {total_items_for_desc} frames. This might be inaccurate. Proceeding...")
1060
+ # If it's 0, the progress in _process_input_source_frames might behave unexpectedly.
1061
+ # Setting to 1 to avoid division by zero, but this means progress won't be very useful.
1062
+ total_items_for_desc = 1 # Fallback to prevent division by zero
1063
+
1064
+ # processed_count_in_provider = 0 # Counts *sampled* frames, not used for progress index
1065
+ last_processed_ms = -float('inf')
1066
+ raw_frames_read_by_provider = 0 # Counts all frames read by cap.read()
1067
+
1068
+ try:
1069
+ while True:
1070
+ # For progress, use current_pos_ms or CAP_PROP_POS_FRAMES
1071
+ # CAP_PROP_POS_FRAMES is a 0-based index of the next frame to be decoded/captured.
1072
+ current_raw_frame_index = int(cap.get(cv2.CAP_PROP_POS_FRAMES)) # Use this for progress
1073
+ current_pos_ms_in_provider = cap.get(cv2.CAP_PROP_POS_MSEC)
1074
+
1075
+ # Loop break condition (more robust)
1076
+ if raw_frames_read_by_provider > 0 and current_pos_ms_in_provider <= last_processed_ms and interval_ms > 0 :
1077
+ # If interval_ms is 0 or very small, current_pos_ms might not advance much for consecutive reads.
1078
+ # Adding a check for raw_frames_read_by_provider against a large number or CAP_PROP_FRAME_COUNT
1079
+ # could be an additional safety, but CAP_PROP_FRAME_COUNT can be unreliable.
1080
+ # The ret_frame check is the primary exit.
1081
+ pass # Let ret_frame handle the actual end. This check is for stuck videos.
1082
+
1083
+
1084
+ should_process_this_frame = current_pos_ms_in_provider >= last_processed_ms + interval_ms - 1
1085
+
1086
+ ret_frame, frame_cv_data = cap.read()
1087
+ if not ret_frame: # Primary exit point for the loop
1088
+ break
1089
+ raw_frames_read_by_provider +=1 # Incremented after successful read
1090
+
1091
+ if should_process_this_frame:
1092
+ try:
1093
+ pil_img = convert_to_pil(frame_cv_data)
1094
+ last_processed_ms = current_pos_ms_in_provider
1095
+ yield pil_img, int(current_pos_ms_in_provider), current_raw_frame_index + 1, total_items_for_desc # Yield 1-based raw frame index
1096
+ except Exception as e_conv:
1097
+ print(f"Error converting frame at {current_pos_ms_in_provider}ms (raw index {current_raw_frame_index}) for {video_original_filename}: {e_conv}. Skipping.")
1098
+ finally:
1099
+ pass
1100
+ finally:
1101
+ if cap.isOpened():
1102
+ cap.release()
1103
+ print(f" Video capture for provider ({video_original_filename}) released.")
1104
+
1105
+ def video_progress_updater(item_progress_value: float, desc: str):
1106
+ overall_progress = (tasks_completed_count + item_progress_value) / total_processing_tasks
1107
+ progress(overall_progress, desc=f"{progress_description_prefix}: {desc}")
1108
+
1109
+ try:
1110
+ zip_file_path_single, status_message_single = _process_input_source_frames(
1111
+ source_file_prefix=video_source_file_prefix,
1112
+ frames_provider=video_frames_provider_generator(video_path_temp, sample_interval_ms),
1113
+ is_video_source=True,
1114
+ enable_person_detection=enable_person_detection,
1115
+ min_target_width_person_percentage=min_target_width_person_percentage,
1116
+ person_model_name=person_model_name,
1117
+ person_conf_threshold=person_conf_threshold,
1118
+ person_iou_threshold=person_iou_threshold,
1119
+ enable_halfbody_detection=enable_halfbody_detection,
1120
+ enable_halfbody_cropping=enable_halfbody_cropping,
1121
+ min_target_width_halfbody_percentage=min_target_width_halfbody_percentage,
1122
+ halfbody_model_name=halfbody_model_name,
1123
+ halfbody_conf_threshold=halfbody_conf_threshold,
1124
+ halfbody_iou_threshold=halfbody_iou_threshold,
1125
+ enable_head_detection=enable_head_detection,
1126
+ enable_head_cropping=enable_head_cropping,
1127
+ min_crop_width_head_percentage=min_crop_width_head_percentage,
1128
+ enable_head_filtering=enable_head_filtering,
1129
+ head_model_name=head_model_name,
1130
+ head_conf_threshold=head_conf_threshold,
1131
+ head_iou_threshold=head_iou_threshold,
1132
+ enable_face_detection=enable_face_detection,
1133
+ enable_face_cropping=enable_face_cropping,
1134
+ min_crop_width_face_percentage=min_crop_width_face_percentage,
1135
+ enable_face_filtering=enable_face_filtering,
1136
+ face_model_name=face_model_name,
1137
+ face_conf_threshold=face_conf_threshold,
1138
+ face_iou_threshold=face_iou_threshold,
1139
+ enable_ccip_classification=enable_ccip_classification,
1140
+ ccip_model_name=ccip_model_name,
1141
+ ccip_threshold=ccip_threshold,
1142
+ enable_lpips_clustering=enable_lpips_clustering,
1143
+ lpips_threshold=lpips_threshold,
1144
+ enable_aesthetic_analysis=enable_aesthetic_analysis,
1145
+ aesthetic_model_name=aesthetic_model_name,
1146
+ progress_updater=video_progress_updater
1147
+ )
1148
+ if zip_file_path_single:
1149
+ output_zip_paths_all_sources.append(zip_file_path_single)
1150
+ all_status_messages.append(f"--- Video ({video_original_filename}) Processing Succeeded ---\n{status_message_single}")
1151
+ else:
1152
+ all_status_messages.append(f"--- Video ({video_original_filename}) Processing Failed ---\n{status_message_single}")
1153
+
1154
+ except Exception as e_vid:
1155
+ # This catches errors if process_video itself raises an unhandled exception
1156
+ # (though process_video has its own try-except)
1157
+ error_msg = f"Critical error during processing of video {video_original_filename}: {e_vid}"
1158
+ print(error_msg)
1159
+ traceback.print_exc()
1160
+ all_status_messages.append(f"--- Video ({video_original_filename}) Processing CRITICALLY FAILED ---\n{error_msg}")
1161
+
1162
+ tasks_completed_count += 1
1163
+ print(f">>> Finished attempt for Video: {video_original_filename} <<<")
1164
+ # Gradio manages the lifecycle of video_path_temp (the uploaded temp file)
1165
+
1166
+ final_summary_message = "\n\n==============================\n\n".join(all_status_messages)
1167
+
1168
+ successful_zips_count = len(output_zip_paths_all_sources)
1169
+ if successful_zips_count == 0 and total_processing_tasks > 0:
1170
+ final_summary_message = f"ALL {total_processing_tasks} INPUT SOURCE(S) FAILED TO PRODUCE A ZIP FILE.\n\n" + final_summary_message
1171
+ elif total_processing_tasks > 0:
1172
+ final_summary_message = f"Successfully processed {successful_zips_count} out of {total_processing_tasks} input source(s).\n\n" + final_summary_message
1173
+ else: # Should be caught earlier by "No processable files"
1174
+ final_summary_message = "No inputs were processed."
1175
+
1176
+ progress(1.0, desc="All processing attempts finished.")
1177
+
1178
+ # gr.Files output expects a list of file paths. An empty list is fine if no files.
1179
+ return output_zip_paths_all_sources, final_summary_message
1180
+
1181
+
1182
+ # --- Gradio Interface Setup ---
1183
+
1184
+ css = """
1185
+ /* Default (Light Mode) Styles */
1186
+ #warning {
1187
+ background-color: #FFCCCB; /* Light red background */
1188
+ padding: 10px;
1189
+ border-radius: 5px;
1190
+ color: #A00000; /* Dark red text */
1191
+ border: 1px solid #E5B8B7; /* A slightly darker border for more definition */
1192
+ }
1193
+ /* Dark Mode Styles */
1194
+ @media (prefers-color-scheme: dark) {
1195
+ #warning {
1196
+ background-color: #5C1A1A; /* Darker red background, suitable for dark mode */
1197
+ color: #FFDDDD; /* Light pink text, for good contrast against the dark red background */
1198
+ border: 1px solid #8B0000; /* A more prominent dark red border in dark mode */
1199
+ }
1200
+ }
1201
+ #status_box {
1202
+ white-space: pre-wrap !important; /* Ensure status messages show newlines */
1203
+ font-family: monospace; /* Optional: Use monospace for better alignment */
1204
+ }
1205
+ """
1206
+
1207
+ # --- Define Model Lists ---
1208
+ person_models = ['person_detect_v1.3_s', 'person_detect_v1.2_s', 'person_detect_v1.1_s', 'person_detect_v1.1_m', 'person_detect_v1_m', 'person_detect_v1.1_n', 'person_detect_v0_s', 'person_detect_v0_m', 'person_detect_v0_x']
1209
+ halfbody_models = ['halfbody_detect_v1.0_s', 'halfbody_detect_v1.0_n', 'halfbody_detect_v0.4_s', 'halfbody_detect_v0.3_s', 'halfbody_detect_v0.2_s']
1210
+ head_models = ['head_detect_v2.0_s', 'head_detect_v2.0_m', 'head_detect_v2.0_n', 'head_detect_v2.0_x', 'head_detect_v2.0_s_yv11', 'head_detect_v2.0_m_yv11', 'head_detect_v2.0_n_yv11', 'head_detect_v2.0_x_yv11', 'head_detect_v2.0_l_yv11']
1211
+ face_models = ['face_detect_v1.4_s', 'face_detect_v1.4_n', 'face_detect_v1.3_s', 'face_detect_v1.3_n', 'face_detect_v1.2_s', 'face_detect_v1.1_s', 'face_detect_v1.1_n', 'face_detect_v1_s', 'face_detect_v1_n', 'face_detect_v0_s', 'face_detect_v0_n']
1212
+ ccip_models = ['ccip-caformer-24-randaug-pruned', 'ccip-caformer-6-randaug-pruned_fp32', 'ccip-caformer-5_fp32']
1213
+ aesthetic_models = ['swinv2pv3_v0_448_ls0.2_x', 'swinv2pv3_v0_448_ls0.2', 'caformer_s36_v0_ls0.2']
1214
+
1215
+ with gr.Blocks(css=css) as demo:
1216
+ gr.Markdown("# Video Processor using dghs-imgutils")
1217
+ gr.Markdown("Upload one or more videos, or a sequence of images. Videos are processed individually, while multiple images are treated as a single sequence. Each processed source (video or image sequence) is then sequentially analyzed by [dghs-imgutils](https://github.com/deepghs/imgutils) to detect subjects, classify items, and process its content according to your settings, ultimately generating a ZIP file with the extracted images.")
1218
+ gr.Markdown("**Detection Flow:** " +
1219
+ "[Person](https://dghs-imgutils.deepghs.org/main/api_doc/detect/person.html) ⇒ " +
1220
+ "[Half-Body](https://dghs-imgutils.deepghs.org/main/api_doc/detect/halfbody.html) (if no person) ⇒ " +
1221
+ "[Face](https://dghs-imgutils.deepghs.org/main/api_doc/detect/face.html) (on target) ⇒ " +
1222
+ "[Head](https://dghs-imgutils.deepghs.org/main/api_doc/detect/head.html) (on target).")
1223
+ gr.Markdown("**Analysis Flow:** " +
1224
+ "[CCIP](https://dghs-imgutils.deepghs.org/main/api_doc/metrics/ccip.html) Clustering ⇒ " +
1225
+ "[LPIPS](https://dghs-imgutils.deepghs.org/main/api_doc/metrics/lpips.html) Clustering ⇒ " +
1226
+ "[Aesthetic](https://dghs-imgutils.deepghs.org/main/api_doc/metrics/dbaesthetic.html) Labeling.")
1227
+ gr.Markdown("**Note on CCIP Folders:** CCIP cluster folders are named `{source_prefix}_ccip_XXX`, sorted by image count (most images = `_ccip_000`).")
1228
+ gr.Markdown("**Note on LPIPS Folders:** LPIPS cluster folders (e.g., `lpips_XXX` or `lpips_sub_XXX`) are also sorted by image count within their scope. 'noise' folders are named explicitly.")
1229
+
1230
+ with gr.Row():
1231
+ with gr.Column(scale=1):
1232
+ # --- Input Components ---
1233
+ process_button = gr.Button("Process Input(s) & Generate ZIP(s)", variant="primary")
1234
+ input_files = gr.Files(label="Upload Videos or Image Sequences", file_types=['video', 'image'], file_count="multiple")
1235
+ sample_interval_ms = gr.Number(label="Sample Interval (ms, for videos)", value=1000, minimum=1, step=100)
1236
+
1237
+ # --- Detection Options ---
1238
+ gr.Markdown("**Detection Options**")
1239
+ # --- Person Detection Block ---
1240
+ with gr.Accordion("Person Detection Options", open=True):
1241
+ enable_person_detection = gr.Checkbox(label="Enable Person Detection", value=True)
1242
+ with gr.Group() as person_detection_params_group:
1243
+ min_target_width_person_percentage_slider = gr.Slider(
1244
+ minimum=0.0, maximum=1.0, value=0.25, step=0.01,
1245
+ label="Min Target Width (% of Item Width)",
1246
+ info="Minimum width for a detected person to be processed (e.g., 0.25 = 25%)."
1247
+ )
1248
+ person_model_name_dd = gr.Dropdown(person_models, label="PD Model", value=person_models[0])
1249
+ person_conf_threshold = gr.Slider(0.0, 1.0, value=0.3, step=0.05, label="PD Conf")
1250
+ person_iou_threshold = gr.Slider(0.0, 1.0, value=0.5, step=0.05, label="PD IoU")
1251
+ enable_person_detection.change(fn=lambda e: gr.update(visible=e), inputs=enable_person_detection, outputs=person_detection_params_group)
1252
+
1253
+ # --- Half-Body Detection Block ---
1254
+ with gr.Accordion("Half-Body Detection Options", open=True):
1255
+ enable_halfbody_detection = gr.Checkbox(label="Enable Half-Body Detection", value=True)
1256
+ with gr.Group() as halfbody_params_group:
1257
+ gr.Markdown("<small>_Detects half-bodies in full items if Person Detection is off/fails._</small>")
1258
+ enable_halfbody_cropping = gr.Checkbox(label="Use Half-Bodies as Targets", value=True)
1259
+ min_target_width_halfbody_percentage_slider = gr.Slider(
1260
+ minimum=0.0, maximum=1.0, value=0.25, step=0.01,
1261
+ label="Min Target Width (% of Item Width)",
1262
+ info="Minimum width for a detected half-body to be processed (e.g., 0.25 = 25%)."
1263
+ )
1264
+ halfbody_model_name_dd = gr.Dropdown(halfbody_models, label="HBD Model", value=halfbody_models[0])
1265
+ halfbody_conf_threshold = gr.Slider(0.0, 1.0, value=0.5, step=0.05, label="HBD Conf")
1266
+ halfbody_iou_threshold = gr.Slider(0.0, 1.0, value=0.7, step=0.05, label="HBD IoU")
1267
+ enable_halfbody_detection.change(fn=lambda e: gr.update(visible=e), inputs=enable_halfbody_detection, outputs=halfbody_params_group)
1268
+
1269
+ # --- Face Detection Block ---
1270
+ with gr.Accordion("Face Detection Options", open=True):
1271
+ enable_face_detection = gr.Checkbox(label="Enable Face Detection", value=True)
1272
+ with gr.Group() as face_params_group:
1273
+ enable_face_filtering = gr.Checkbox(label="Filter Targets Without Detected Faces", value=True)
1274
+ enable_face_cropping = gr.Checkbox(label="Crop Detected Faces", value=False)
1275
+ min_crop_width_face_percentage_slider = gr.Slider(
1276
+ minimum=0.0, maximum=1.0, value=0.2, step=0.01,
1277
+ label="Min Crop Width (% of Parent Width)",
1278
+ info="Minimum width for a face crop relative to its parent image's width (e.g., 0.2 = 20%)."
1279
+ )
1280
+ face_model_name_dd = gr.Dropdown(face_models, label="FD Model", value=face_models[0])
1281
+ face_conf_threshold = gr.Slider(0.0, 1.0, value=0.25, step=0.05, label="FD Conf")
1282
+ face_iou_threshold = gr.Slider(0.0, 1.0, value=0.7, step=0.05, label="FD IoU")
1283
+ enable_face_detection.change(fn=lambda e: gr.update(visible=e), inputs=enable_face_detection, outputs=face_params_group)
1284
+
1285
+ # --- Head Detection Block ---
1286
+ with gr.Accordion("Head Detection Options", open=True):
1287
+ enable_head_detection = gr.Checkbox(label="Enable Head Detection", value=True)
1288
+ with gr.Group() as head_params_group:
1289
+ gr.Markdown("<small>_Detects heads in targets. Crops if meets width req._</small>")
1290
+ enable_head_filtering = gr.Checkbox(label="Filter Targets Without Heads", value=True)
1291
+ enable_head_cropping = gr.Checkbox(label="Crop Detected Heads", value=False)
1292
+ min_crop_width_head_percentage_slider = gr.Slider(
1293
+ minimum=0.0, maximum=1.0, value=0.2, step=0.01,
1294
+ label="Min Crop Width (% of Parent Width)",
1295
+ info="Minimum width for a head crop relative to its parent image's width (e.g., 0.2 = 20%)."
1296
+ )
1297
+ head_model_name_dd = gr.Dropdown(head_models, label="HD Model", value=head_models[0])
1298
+ head_conf_threshold = gr.Slider(0.0, 1.0, value=0.4, step=0.05, label="HD Conf")
1299
+ head_iou_threshold = gr.Slider(0.0, 1.0, value=0.7, step=0.05, label="HD IoU")
1300
+ enable_head_detection.change(fn=lambda e: gr.update(visible=e), inputs=enable_head_detection, outputs=head_params_group)
1301
+
1302
+ # --- Analysis/Classification Options ---
1303
+ gr.Markdown("**Analysis & Classification**")
1304
+ # --- CCIP Classification Block ---
1305
+ with gr.Accordion("CCIP Classification Options", open=True):
1306
+ enable_ccip_classification = gr.Checkbox(label="Enable CCIP Classification", value=True)
1307
+ with gr.Group() as ccip_params_group:
1308
+ gr.Markdown("<small>_Clusters results by similarity. Folders sorted by image count._</small>")
1309
+ ccip_model_name_dd = gr.Dropdown(ccip_models, label="CCIP Model", value=ccip_models[0])
1310
+ ccip_threshold_slider = gr.Slider(0.0, 1.0, step=0.01, value=0.20, label="CCIP Similarity Threshold")
1311
+ enable_ccip_classification.change(fn=lambda e: gr.update(visible=e), inputs=enable_ccip_classification, outputs=ccip_params_group)
1312
+
1313
+ # LPIPS Clustering Options
1314
+ with gr.Accordion("LPIPS Clustering Options", open=True):
1315
+ enable_lpips_clustering = gr.Checkbox(label="Enable LPIPS Clustering", value=True)
1316
+ with gr.Group() as lpips_params_group:
1317
+ gr.Markdown("<small>_Clusters images by LPIPS similarity. Applied after CCIP (if enabled) or globally. Folders sorted by image count._</small>")
1318
+ lpips_threshold_slider = gr.Slider(0.0, 1.0, step=0.01, value=0.45, label="LPIPS Similarity Threshold")
1319
+ enable_lpips_clustering.change(fn=lambda e: gr.update(visible=e), inputs=enable_lpips_clustering, outputs=lpips_params_group)
1320
+
1321
+ # --- Aesthetic Analysis Block ---
1322
+ with gr.Accordion("Aesthetic Analysis Options", open=True):
1323
+ enable_aesthetic_analysis = gr.Checkbox(label="Enable Aesthetic Analysis (Anime)", value=True)
1324
+ with gr.Group() as aesthetic_params_group:
1325
+ gr.Markdown("<small>_Prepends aesthetic label to filenames._</small>")
1326
+ aesthetic_model_name_dd = gr.Dropdown(aesthetic_models, label="Aesthetic Model", value=aesthetic_models[0])
1327
+ enable_aesthetic_analysis.change(fn=lambda e: gr.update(visible=e), inputs=enable_aesthetic_analysis, outputs=aesthetic_params_group)
1328
+
1329
+ gr.Markdown("---")
1330
+ gr.Markdown("**Warning:** Complex combinations can be slow. Models downloaded on first use.", elem_id="warning")
1331
+
1332
+ with gr.Column(scale=1):
1333
+ # --- Output Components ---
1334
+ status_text = gr.Textbox(label="Processing Status", interactive=False, lines=20, elem_id="status_box")
1335
+ output_zips = gr.Files(label="Download Processed Images (ZIPs)")
1336
+
1337
+ # Connect button click
1338
+ process_button.click(
1339
+ fn=process_inputs_main,
1340
+ inputs=[
1341
+ input_files, sample_interval_ms,
1342
+ # Person Detect
1343
+ enable_person_detection, min_target_width_person_percentage_slider,
1344
+ person_model_name_dd, person_conf_threshold, person_iou_threshold,
1345
+ # HalfBody Detect
1346
+ enable_halfbody_detection, enable_halfbody_cropping, min_target_width_halfbody_percentage_slider,
1347
+ halfbody_model_name_dd, halfbody_conf_threshold, halfbody_iou_threshold,
1348
+ # Head Detect
1349
+ enable_head_detection, enable_head_cropping, min_crop_width_head_percentage_slider,
1350
+ enable_head_filtering, head_model_name_dd, head_conf_threshold, head_iou_threshold,
1351
+ # Face Detect
1352
+ enable_face_detection, enable_face_cropping, min_crop_width_face_percentage_slider,
1353
+ enable_face_filtering, face_model_name_dd, face_conf_threshold, face_iou_threshold,
1354
+ # CCIP
1355
+ enable_ccip_classification, ccip_model_name_dd, ccip_threshold_slider,
1356
+ # LPIPS
1357
+ enable_lpips_clustering, lpips_threshold_slider,
1358
+ # Aesthetic
1359
+ enable_aesthetic_analysis, aesthetic_model_name_dd,
1360
+ ],
1361
+ outputs=[output_zips, status_text]
1362
+ )
1363
+
1364
+ # --- Launch Script ---
1365
+ if __name__ == "__main__":
1366
+ print("Starting Gradio App...")
1367
+ # Model pre-check
1368
+ try:
1369
+ print("Checking/Downloading models (this might take a moment)...")
1370
+ # Use simple, small images for checks
1371
+ dummy_img_pil = Image.new('RGB', (64, 64), color = 'orange')
1372
+ print(" - Person detection...")
1373
+ _ = person_detector.detect_person(dummy_img_pil, model_name=person_models[0])
1374
+ print(" - HalfBody detection...")
1375
+ _ = halfbody_detector.detect_halfbody(dummy_img_pil, model_name=halfbody_models[0])
1376
+ print(" - Head detection...")
1377
+ _ = head_detector.detect_heads(dummy_img_pil, model_name=head_models[0])
1378
+ print(" - Face detection...")
1379
+ _ = face_detector.detect_faces(dummy_img_pil, model_name=face_models[0])
1380
+ print(" - CCIP feature extraction...")
1381
+ _ = ccip_analyzer.ccip_extract_feature(dummy_img_pil, size=384, model=ccip_models[0])
1382
+ print(" - LPIPS feature extraction...")
1383
+ _ = lpips_module.lpips_extract_feature(dummy_img_pil)
1384
+ print(" - Aesthetic analysis...")
1385
+ _ = dbaesthetic_analyzer.anime_dbaesthetic(dummy_img_pil, model_name=aesthetic_models[0])
1386
+ print("Models seem ready or downloaded.")
1387
+ del dummy_img_pil
1388
+ gc.collect()
1389
+ except Exception as model_err:
1390
+ print(f"\n--- !!! WARNING !!! ---")
1391
+ print(f"Could not pre-check/download all models: {model_err}")
1392
+ print(f"Models will be downloaded when first used by the application, which may cause a delay on the first run.")
1393
+ print(f"Check your internet connection and library installation (pip install \"dghs-imgutils[gpu]\").")
1394
+ print(f"-----------------------\n")
1395
+ # Launch the app
1396
+ demo.launch(inbrowser=True)
requirements.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ gradio==5.29.0
2
+ dghs-imgutils[gpu]
webui.bat ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ @echo off
2
+
3
+ :: The source of the webui.bat file is stable-diffusion-webui
4
+ :: set COMMANDLINE_ARGS=--whisper_implementation faster-whisper --input_audio_max_duration -1 --default_model_name large-v2 --auto_parallel True --output_dir output --vad_max_merge_size 90 --save_downloaded_files --autolaunch
5
+
6
+ if not defined PYTHON (set PYTHON=python)
7
+ if not defined VENV_DIR (set "VENV_DIR=%~dp0%venv")
8
+
9
+ mkdir tmp 2>NUL
10
+
11
+ %PYTHON% -c "" >tmp/stdout.txt 2>tmp/stderr.txt
12
+ if %ERRORLEVEL% == 0 goto :check_pip
13
+ echo Couldn't launch python
14
+ goto :show_stdout_stderr
15
+
16
+ :check_pip
17
+ %PYTHON% -mpip --help >tmp/stdout.txt 2>tmp/stderr.txt
18
+ if %ERRORLEVEL% == 0 goto :start_venv
19
+ if "%PIP_INSTALLER_LOCATION%" == "" goto :show_stdout_stderr
20
+ %PYTHON% "%PIP_INSTALLER_LOCATION%" >tmp/stdout.txt 2>tmp/stderr.txt
21
+ if %ERRORLEVEL% == 0 goto :start_venv
22
+ echo Couldn't install pip
23
+ goto :show_stdout_stderr
24
+
25
+ :start_venv
26
+ if ["%VENV_DIR%"] == ["-"] goto :skip_venv
27
+ if ["%SKIP_VENV%"] == ["1"] goto :skip_venv
28
+
29
+ dir "%VENV_DIR%\Scripts\Python.exe" >tmp/stdout.txt 2>tmp/stderr.txt
30
+ if %ERRORLEVEL% == 0 goto :activate_venv
31
+
32
+ for /f "delims=" %%i in ('CALL %PYTHON% -c "import sys; print(sys.executable)"') do set PYTHON_FULLNAME="%%i"
33
+ echo Creating venv in directory %VENV_DIR% using python %PYTHON_FULLNAME%
34
+ %PYTHON_FULLNAME% -m venv "%VENV_DIR%" >tmp/stdout.txt 2>tmp/stderr.txt
35
+ if %ERRORLEVEL% == 0 goto :activate_venv
36
+ echo Unable to create venv in directory "%VENV_DIR%"
37
+ goto :show_stdout_stderr
38
+
39
+ :activate_venv
40
+ set PYTHON="%VENV_DIR%\Scripts\Python.exe"
41
+ echo venv %PYTHON%
42
+
43
+ :skip_venv
44
+ goto :launch
45
+
46
+ :launch
47
+ %PYTHON% app.py %COMMANDLINE_ARGS% %*
48
+ pause
49
+ exit /b
50
+
51
+ :show_stdout_stderr
52
+
53
+ echo.
54
+ echo exit code: %errorlevel%
55
+
56
+ for /f %%i in ("tmp\stdout.txt") do set size=%%~zi
57
+ if %size% equ 0 goto :show_stderr
58
+ echo.
59
+ echo stdout:
60
+ type tmp\stdout.txt
61
+
62
+ :show_stderr
63
+ for /f %%i in ("tmp\stderr.txt") do set size=%%~zi
64
+ if %size% equ 0 goto :show_stderr
65
+ echo.
66
+ echo stderr:
67
+ type tmp\stderr.txt
68
+
69
+ :endofscript
70
+
71
+ echo.
72
+ echo Launch unsuccessful. Exiting.
73
+ pause