Meehai commited on
Commit
15a0f2e
·
2 Parent(s): 4e44b63 d05f7e1

Merge branch 'main' of hf.co:datasets/Meehai/dronescapes

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. README.md +27 -156
  2. collage.png +2 -2
  3. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4501.npz +0 -3
  4. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4502.npz +0 -3
  5. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4503.npz +0 -3
  6. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4504.npz +0 -3
  7. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4505.npz +0 -3
  8. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4506.npz +0 -3
  9. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4507.npz +0 -3
  10. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4508.npz +0 -3
  11. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4509.npz +0 -3
  12. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4510.npz +0 -3
  13. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4511.npz +0 -3
  14. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4512.npz +0 -3
  15. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4513.npz +0 -3
  16. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4514.npz +0 -3
  17. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4515.npz +0 -3
  18. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4516.npz +0 -3
  19. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4517.npz +0 -3
  20. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4518.npz +0 -3
  21. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4519.npz +0 -3
  22. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4520.npz +0 -3
  23. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4521.npz +0 -3
  24. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4522.npz +0 -3
  25. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4523.npz +0 -3
  26. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4524.npz +0 -3
  27. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4525.npz +0 -3
  28. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4526.npz +0 -3
  29. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4527.npz +0 -3
  30. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4528.npz +0 -3
  31. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4529.npz +0 -3
  32. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4530.npz +0 -3
  33. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4531.npz +0 -3
  34. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4532.npz +0 -3
  35. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4533.npz +0 -3
  36. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4534.npz +0 -3
  37. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4535.npz +0 -3
  38. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4536.npz +0 -3
  39. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4537.npz +0 -3
  40. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4538.npz +0 -3
  41. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4539.npz +0 -3
  42. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4540.npz +0 -3
  43. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4541.npz +0 -3
  44. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4542.npz +0 -3
  45. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4543.npz +0 -3
  46. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4544.npz +0 -3
  47. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4545.npz +0 -3
  48. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4546.npz +0 -3
  49. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4547.npz +0 -3
  50. data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4548.npz +0 -3
README.md CHANGED
@@ -1,162 +1,36 @@
 
 
 
 
 
 
 
 
 
1
  # Dronescapes dataset
2
 
3
  As introduced in our ICCV 2023 workshop paper: [link](https://openaccess.thecvf.com/content/ICCV2023W/LIMIT/papers/Marcu_Self-Supervised_Hypergraphs_for_Learning_Multiple_World_Interpretations_ICCVW_2023_paper.pdf)
4
 
5
  ![Logo](logo.png)
6
 
7
- # 1. Downloading the data
 
8
 
9
- ## Option 1. Download the pre-processed dataset from HuggingFace repository
10
 
11
  ```
12
  git lfs install # Make sure you have git-lfs installed (https://git-lfs.com)
13
  git clone https://huggingface.co/datasets/Meehai/dronescapes
14
  ```
15
 
16
- Note: the dataset has about 500GB, so it may take a while to clone it.
17
-
18
- <details>
19
- <summary> <b> Option 2. Generating the dataset from raw videos and basic labels </b>.</summary>
20
-
21
- Recommended if you intend on understanding how the dataset was created or add new videos or representations.
22
-
23
- ### 1.2.1 Raw videos
24
-
25
- Follow the commands in each directory under `raw_data/videos/*/commands.txt` if you want to start from the 4K videos.
26
-
27
- If you only want the 540p videos as used in the paper, they are already provided in the `raw_data/videos/*` directories.
28
-
29
- ### 1.2.2 Semantic segmentation labels (human annotated)
30
-
31
- These were human annotated and then propagated using [segprop](https://github.com/vlicaret/segprop).
32
-
33
- ```bash
34
- cd raw_data/
35
- tar -xzvf segprop_npz_540.tar.gz
36
- ```
37
-
38
- ### 1.2.3 Generate the rest of the representations
39
-
40
- We use the [video-representations-extractor](https://gitlab.com/meehai/video-representations-extractor) to generate
41
- the rest of the labels using pre-traing networks or algoritms.
42
-
43
- Install it via `pip install video-representations-extractor` (or follow the README over there for docker or local env)
44
-
45
- ```
46
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=0 vre raw_data/videos/atanasie_DJI_0652_full/atanasie_DJI_0652_full_540p.mp4 -o raw_data/npz_540p/atanasie_DJI_0652_full/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
47
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=1 vre raw_data/videos/barsana_DJI_0500_0501_combined_sliced_2700_14700/barsana_DJI_0500_0501_combined_sliced_2700_14700_540p.mp4 -o raw_data/npz_540p/barsana_DJI_0500_0501_combined_sliced_2700_14700/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
48
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=2 vre raw_data/videos/comana_DJI_0881_full/comana_DJI_0881_full_540p.mp4 -o raw_data/npz_540p/comana_DJI_0881_full/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
49
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=3 vre raw_data/videos/gradistei_DJI_0787_0788_0789_combined_sliced_3510_13110/gradistei_DJI_0787_0788_0789_combined_sliced_3510_13110_540p.mp4 -o raw_data/npz_540p/gradistei_DJI_0787_0788_0789_combined_sliced_3510_13110/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
50
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=4 vre raw_data/videos/herculane_DJI_0021_full/herculane_DJI_0021_full_540p.mp4 -o raw_data/npz_540p/herculane_DJI_0021_full/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
51
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=5 vre raw_data/videos/jupiter_DJI_0703_0704_0705_combined_sliced_10650_21715/jupiter_DJI_0703_0704_0705_combined_sliced_10650_21715_540p.mp4 -o raw_data/npz_540p/jupiter_DJI_0703_0704_0705_combined_sliced_10650_21715/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
52
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=6 vre raw_data/videos/norway_210821_DJI_0015_full/norway_210821_DJI_0015_full_540p.mp4 -o raw_data/npz_540p/norway_210821_DJI_0015_full/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
53
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=7 vre raw_data/videos/olanesti_DJI_0416_full/olanesti_DJI_0416_full_540p.mp4 -o raw_data/npz_540p/olanesti_DJI_0416_full/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
54
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=0 vre raw_data/videos/petrova_DJI_0525_0526_combined_sliced_2850_11850/petrova_DJI_0525_0526_combined_sliced_2850_11850_540p.mp4 -o raw_data/npz_540p/petrova_DJI_0525_0526_combined_sliced_2850_11850/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
55
- VRE_DEVICE=cuda CUDA_VISIBLE_DEVICES=1 vre raw_data/videos/slanic_DJI_0956_0957_combined_sliced_780_9780/slanic_DJI_0956_0957_combined_sliced_780_9780_540p.mp4 -o raw_data/npz_540p/slanic_DJI_0956_0957_combined_sliced_780_9780/ --cfg_path scripts/cfg.yaml --batch_size 3 --n_threads_data_storer 4 --output_dir_exist_mode overwrite
56
- ```
57
-
58
- Note: `depth_sfm`, `normals_sfm` and `depth_ufo` are not available in VRE. Contact us for more info about them.
59
-
60
- Note: Add `--representations "rgb" "opticalflow_rife" "depth_dpt" "edges_dexined" "semantic_mask2former_swin_mapillary" "softseg_gb"` to control if you only want a subset of the representations.
61
-
62
- Note: Some batch sizes are overwritten in the config itself.
63
-
64
- ### 1.2.4 Convert Mask2Former from Mapillary classes to segprop8 classes
65
-
66
- Since we are using pre-trained Mask2Former which has either mapillary or COCO panoptic classes, we need to convert them to dronescapes-compatible (8) classes.
67
-
68
- To do this, we use the `scripts/convert_m2f_to_dronescapes.py` script:
69
- ```
70
- python scripts/convert_m2f_to_dronescapes.py in_dir out_dir mapillary/coco [--overwrite]
71
- ```
72
-
73
- ```
74
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/atanasie_DJI_0652_full/semantic_mask2former_swin_mapillary raw_data/npz_540p/atanasie_DJI_0652_full/semantic_mask2former_swin_mapillary_converted mapillary
75
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/barsana_DJI_0500_0501_combined_sliced_2700_14700/semantic_mask2former_swin_mapillary raw_data/npz_540p/barsana_DJI_0500_0501_combined_sliced_2700_14700/semantic_mask2former_swin_mapillary_converted mapillary
76
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/comana_DJI_0881_full/semantic_mask2former_swin_mapillary raw_data/npz_540p/comana_DJI_0881_full/semantic_mask2former_swin_mapillary_converted mapillary
77
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/gradistei_DJI_0787_0788_0789_combined_sliced_3510_13110/semantic_mask2former_swin_mapillary raw_data/npz_540p/gradistei_DJI_0787_0788_0789_combined_sliced_3510_13110/semantic_mask2former_swin_mapillary_converted mapillary
78
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/herculane_DJI_0021_full/semantic_mask2former_swin_mapillary raw_data/npz_540p/herculane_DJI_0021_full/semantic_mask2former_swin_mapillary_converted mapillary
79
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/jupiter_DJI_0703_0704_0705_combined_sliced_10650_21715/semantic_mask2former_swin_mapillary raw_data/npz_540p/jupiter_DJI_0703_0704_0705_combined_sliced_10650_21715/semantic_mask2former_swin_mapillary_converted mapillary
80
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/norway_210821_DJI_0015_full/semantic_mask2former_swin_mapillary raw_data/npz_540p/norway_210821_DJI_0015_full/semantic_mask2former_swin_mapillary_converted mapillary
81
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/olanesti_DJI_0416_full/semantic_mask2former_swin_mapillary raw_data/npz_540p/olanesti_DJI_0416_full/semantic_mask2former_swin_mapillary_converted mapillary
82
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/petrova_DJI_0525_0526_combined_sliced_2850_11850/semantic_mask2former_swin_mapillary raw_data/npz_540p/petrova_DJI_0525_0526_combined_sliced_2850_11850/semantic_mask2former_swin_mapillary_converted mapillary
83
- python scripts/convert_m2f_to_dronescapes.py raw_data/npz_540p/slanic_DJI_0956_0957_combined_sliced_780_9780/semantic_mask2former_swin_mapillary raw_data/npz_540p/slanic_DJI_0956_0957_combined_sliced_780_9780/semantic_mask2former_swin_mapillary_converted mapillary
84
- ```
85
-
86
- ### 1.2.5 Check counts for consistency
87
-
88
- Run: `bash scripts/count_npz.sh raw_data/npz_540p`. At this point it should return:
89
- | scene | rgb | depth_dpt | depth_sfm_manual20.. | edges_dexined | normals_sfm_manual.. | opticalflow_rife | semantic_mask2form.. | semantic_segprop8 |
90
- |:----------|------:|------------:|-----------------------:|----------------:|-----------------------:|-------------------:|-----------------------:|--------------------:|
91
- | atanasie | 9021 | 9021 | 9020 | 9021 | 9020 | 9021 | 9021 | 9001 |
92
- | barsana | 12001 | 12001 | 12001 | 12001 | 12001 | 12000 | 12001 | 1573 |
93
- | comana | 9022 | 9022 | 0 | 9022 | 0 | 9022 | 9022 | 1210 |
94
- | gradistei | 9601 | 9601 | 9600 | 9601 | 9600 | 9600 | 9601 | 1210 |
95
- | herculane | 9022 | 9022 | 9021 | 9022 | 9021 | 9022 | 9022 | 1210 |
96
- | jupiter | 11066 | 11066 | 11065 | 11066 | 11065 | 11066 | 11066 | 1452 |
97
- | norway | 2983 | 2983 | 0 | 2983 | 0 | 2983 | 2983 | 2941 |
98
- | olanesti | 9022 | 9022 | 9021 | 9022 | 9021 | 9022 | 9022 | 1210 |
99
- | petrova | 9001 | 9001 | 9001 | 9001 | 9001 | 9000 | 9001 | 1210 |
100
- | slanic | 9001 | 9001 | 9001 | 9001 | 9001 | 9000 | 9001 | 9001 |
101
-
102
- ### 1.2.6. Split intro train, validation, semisupervised and train
103
-
104
- We include 8 splits: 4 using only GT annotated semantic data and 4 using all available data (i.e. segproped between
105
- annotated data). The indexes are taken from `txt_files/*`, i.e. `txt_files/manually_adnotated_files/test_files_116.txt`
106
- refers to the fact that the (unseen at train time) test set (norway + petrova + barsana) contains 116 manually
107
- annotated semantic files. We include all representations from above, not just semantic for all possible splits.
108
- Adding new representations is as simple as running VRE on the 540p mp4 file
109
 
110
- ```
111
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/annotated_and_segprop/train_files_11664.txt -o data/train_set --overwrite
112
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/annotated_and_segprop/val_files_605.txt -o data/validation_set --overwrite
113
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/annotated_and_segprop/semisup_files_11299.txt -o data/semisupervised_set --overwrite
114
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/annotated_and_segprop/test_files_5603.txt -o data/test_set --overwrite
115
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/manually_annotated_files/train_files_218.txt -o data/train_set_annotated_only --overwrite
116
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/manually_annotated_files/val_files_15.txt -o data/validation_set_annotated_only --overwrite
117
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/manually_annotated_files/semisup_files_207.txt -o data/semisupervised_set_annotated_nly --overwrite
118
- python scripts/symlinks_from_txt_list.py raw_data/npz_540p/ --txt_file txt_files/manually_annotated_files/test_files_116.txt -o data/test_set_annotated_only --overwrite
119
- ```
120
 
121
- Note: `add --copy_files` if you want to make copies instead of using symlinks.
122
-
123
- Upon calling this, you should be able to see something like this:
124
- ```
125
- user> ls data/*
126
- data/semisupervised_set:
127
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
128
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
129
-
130
- data/semisupervised_set_annotated_nly:
131
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
132
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
133
-
134
- data/test_set:
135
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
136
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
137
-
138
- data/test_set_annotated_nly:
139
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
140
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
141
-
142
- data/train_set:
143
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
144
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
145
-
146
- data/train_set_annotated_only:
147
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
148
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
149
-
150
- data/validation_set:
151
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
152
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
153
-
154
- data/validation_set_annotated_only:
155
- depth_dpt edges_dexined opticalflow_rife semantic_mask2former_swin_mapillary_converted
156
- depth_sfm_manual202204 normals_sfm_manual202204 rgb semantic_segprop8
157
- ```
158
 
159
- ### 1.2.7 Convert Camera Normals to World Normals
160
 
161
  This is an optional step, but for some use cases, it may be better to use world normals instead of camera normals, which
162
  are provided by default in `normals_sfm_manual202204`. To convert, we provide camera rotation matrices in
@@ -172,7 +46,7 @@ def convert_camera_to_world(normals: np.ndarray, rotation_matrix: np.ndarray) ->
172
  return np.clip(camera_normals, 0.0, 1.0)
173
  ```
174
 
175
- ### 1.2.8 Raw camera location
176
 
177
  The GPS location (lat/long/height) is in the `raw_data/raw_camera_info.tar.gz`. Each file over there is an archive over the
178
  original videos, so the train/val/test splits files from `raw_data/txt_files` must be used to map to proper indices. Each
@@ -186,20 +60,16 @@ Note: there is no GPS for the norway scene, the log seems to have been lost.
186
 
187
  As per the split from the paper:
188
 
189
- <details>
190
  <summary> Split </summary>
191
- <img src="split.png">
192
- </details>
193
 
194
- The data is in `data/*` (see the `ls` call above, it should match even if you download from huggingface).
195
 
196
- ## 2.1 Using the provided viewer
 
197
 
198
  ![Collage](collage.png)
199
 
200
- The simplest way to explore the data is to use the [provided notebook](scripts/dronescapes_viewer.ipynb). Upon running
201
- it, you should get a collage with all the default tasks, like the picture at the top.
202
-
203
  For a CLI-only method, you can use the provided reader as well:
204
 
205
  ```
@@ -282,7 +152,9 @@ python scripts/evaluate_semantic_segmentation.py y_dir gt_dir -o results.csv --c
282
  <summary> Script explanation </summary>
283
  The script is a bit convoluted, so let's break it into parts:
284
 
285
- - `y_dir` and `gt_dir` Two directories of .npz files in the same format as the dataset (y_dir/1.npz, gt_dir/55.npz etc.)
 
 
286
  - `classes` A list of classes in the order that they appear in the predictions and gt files
287
  - `class_weights` (optional, but used in paper) How much to weigh each class. In the paper we compute these weights as
288
  the number of pixels in all the dataset (train/val/semisup/test) for each of the 8 classes resulting in the numbers
@@ -298,7 +170,7 @@ outputs one csv file with predictions for each npz file, the scenes are used for
298
 
299
  ```
300
  python scripts/evaluate_semantic_segmentation.py \
301
- data/test_set_annotated_only/semantic_mask2former_swin_mapillary_converted/ \ # change this with your predictions dir
302
  data/test_set_annotated_only/semantic_segprop8/ \
303
  -o results.csv \
304
  --classes land forest residential road little-objects water sky hill \
@@ -343,5 +215,4 @@ all 60.456 73.261
343
 
344
  | method | #paramters | average | barsana_DJI_0500_0501_combined_sliced_2700_14700 | comana_DJI_0881_full | norway_210821_DJI_0015_full |
345
  |:-|:-|:-|:-|:-|:-|
346
- | [Mask2Former](https://openaccess.thecvf.com/content/CVPR2022/papers/Cheng_Masked-Attention_Mask_Transformer_for_Universal_Image_Segmentation_CVPR_2022_paper.pdf) | 216M | 65.01 | 75.33 | 73.77 | 45.93 |
347
-
 
1
+ ---
2
+ tags:
3
+ - videos
4
+ - video
5
+ - uav
6
+ - drones
7
+ - multitask
8
+ - multimodal
9
+ ---
10
  # Dronescapes dataset
11
 
12
  As introduced in our ICCV 2023 workshop paper: [link](https://openaccess.thecvf.com/content/ICCV2023W/LIMIT/papers/Marcu_Self-Supervised_Hypergraphs_for_Learning_Multiple_World_Interpretations_ICCVW_2023_paper.pdf)
13
 
14
  ![Logo](logo.png)
15
 
16
+ Note: We are extending this dataset in another repository. GT data for benchmarking is the same, but we are generating
17
+ modalities as inputs: [dronescapes-2024](https://huggingface.co/datasets/Meehai/dronescapes-2024).
18
 
19
+ # 1. Downloading the data
20
 
21
  ```
22
  git lfs install # Make sure you have git-lfs installed (https://git-lfs.com)
23
  git clone https://huggingface.co/datasets/Meehai/dronescapes
24
  ```
25
 
26
+ Note: the dataset has about 200GB, so it may take a while to clone it.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
27
 
28
+ ## 1.2 Low level data for the dataset (GPS, camera rotation matrices)
 
 
 
 
 
 
 
 
 
29
 
30
+ <details>
31
+ <summary> Click to expand </summary>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
32
 
33
+ ### 1.2.1 Convert Camera Normals to World Normals
34
 
35
  This is an optional step, but for some use cases, it may be better to use world normals instead of camera normals, which
36
  are provided by default in `normals_sfm_manual202204`. To convert, we provide camera rotation matrices in
 
46
  return np.clip(camera_normals, 0.0, 1.0)
47
  ```
48
 
49
+ ### 1.2.2 Raw camera location
50
 
51
  The GPS location (lat/long/height) is in the `raw_data/raw_camera_info.tar.gz`. Each file over there is an archive over the
52
  original videos, so the train/val/test splits files from `raw_data/txt_files` must be used to map to proper indices. Each
 
60
 
61
  As per the split from the paper:
62
 
 
63
  <summary> Split </summary>
64
+ <img src="split.png" width="500px">
 
65
 
66
+ The data is in the `data*` directory with 1 sub-directory for each split above (and a few more variants).
67
 
68
+ The simplest way to explore the data is to use the [provided notebook](scripts/dronescapes_viewer.ipynb). Upon running
69
+ it, you should get a collage with all the default tasks, like this:
70
 
71
  ![Collage](collage.png)
72
 
 
 
 
73
  For a CLI-only method, you can use the provided reader as well:
74
 
75
  ```
 
152
  <summary> Script explanation </summary>
153
  The script is a bit convoluted, so let's break it into parts:
154
 
155
+ - `y_dir` and `gt_dir` Two directories of .npz files in the same format as the dataset
156
+ - y_dir/1.npz, ..., y_dir/N.npz
157
+ - gt_dir/1.npz, ..., gt_dir.npz
158
  - `classes` A list of classes in the order that they appear in the predictions and gt files
159
  - `class_weights` (optional, but used in paper) How much to weigh each class. In the paper we compute these weights as
160
  the number of pixels in all the dataset (train/val/semisup/test) for each of the 8 classes resulting in the numbers
 
170
 
171
  ```
172
  python scripts/evaluate_semantic_segmentation.py \
173
+ data/test_set_annotated_only/semantic_mask2former_swin_mapillary_converted/ \ # Mask2Former example, use yours here!
174
  data/test_set_annotated_only/semantic_segprop8/ \
175
  -o results.csv \
176
  --classes land forest residential road little-objects water sky hill \
 
215
 
216
  | method | #paramters | average | barsana_DJI_0500_0501_combined_sliced_2700_14700 | comana_DJI_0881_full | norway_210821_DJI_0015_full |
217
  |:-|:-|:-|:-|:-|:-|
218
+ | [Mask2Former](https://openaccess.thecvf.com/content/CVPR2022/papers/Cheng_Masked-Attention_Mask_Transformer_for_Universal_Image_Segmentation_CVPR_2022_paper.pdf) | 216M | 65.01 | 75.33 | 73.77 | 45.93 |
 
collage.png CHANGED

Git LFS Details

  • SHA256: 842395b7e91d7a34dbb1d5c18e261ec97ae19f67700a5a716fb042d347a3fd42
  • Pointer size: 132 Bytes
  • Size of remote file: 5.27 MB

Git LFS Details

  • SHA256: 5db9b6bdae025c95b3464fec742315d98c490100e76eb658550763f4ca2178c5
  • Pointer size: 132 Bytes
  • Size of remote file: 3.6 MB
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4501.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:9221634356fa92c5c8d64ac065522fb822899d2a2007ee487ff9d9d84411e040
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4502.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:e8fcc02104731fda2a433606fbc593ccd3923f697d8b6ab75d4c2024c5245dce
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4503.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:087a126cb74df6f45b5a9ab6b32c2f0efc025a13c3a08e14e518d62e510ec1f5
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4504.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:8c825ee2818c72d28259c26747e376470d795f4b9f186b201c03143d036abf8f
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4505.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:143c9b36e0056c82d0ea8a8afea0b7012fd40c494464a2290a7d1c74d9d11c7e
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4506.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:f2890455ff8682efa54e7ddbc24f3b9deeb3700f620000cb3ca5aca743abfba6
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4507.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c58975be5ee346940921aae1bcbed73cbe8b7bee1785d3c0a2c3c6629e8b1be4
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4508.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:9d5c6d952415d9e03d8877c713b04f8bd495a9ed7cb28ef1a3ad99d8e457a1a6
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4509.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:16f280ca6e9d403c9309ec095bb2cc5b0285b1e0847ca19623c2e24441ef6726
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4510.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:70604900589f69913c669105479fa9c16139236511319bad00962b4549282013
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4511.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c523f2788c435a5cc4137a69101194f10783c8346beb22bfe3d2f4f49fa40a58
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4512.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:a0d339a65826aa0f4782590263442e14bbb32c435d48bbd24e27cd4d6b3f99e2
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4513.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c04d1cea70dd6c1ad2bb98e87650280e8ab8356fccd3aac459a228fa9ac9c3a3
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4514.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:6314ceb02fe496c71ba3c1a4dc85d4616e54d3b9a3089c26d9ea401a6a255457
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4515.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:050288de882f08cca86270498a2e1d0c18099735ea09ca40134bd8cd5d1c2df6
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4516.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:74813236ed8cf122378a94cbdea9d6b6241669d257d18e18eee79f69ac801475
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4517.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:ecb19b69b1a86e884ae0ed5ff2af335c081b0cfb0be61ad5050c172bd5450c61
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4518.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:14a2bd02f6b03e6337ce924b46916b4fc1ade7c7724a58ca49fe2b709b1dbdfe
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4519.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:52f7b539e18a90183da5f5f29a62d38e71789c843b58b205cdd87093b609b466
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4520.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c9df6d23995fe806ae3631ac228ef39cc99d2d2a334072227e9ad533dc296064
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4521.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:638f4c5426d521e69e7ff8cc1ab1767d7fbf5c77ac6cb185b37953c0f97f8394
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4522.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:199ba423f3a6f1f2648ca7e1544746bdd30519c029aa4c1b3f26c0eafd0236cf
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4523.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:715246a390dc4e638fe2d512b72842c95004d8a433a75bee20b5cabdcf17c692
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4524.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:a3619319a152ebfaa277e57bf8a563b73f87dad90aaa5ecce0c8e90d42b40dca
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4525.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:f0212335dc0b131c4a8c6ad99ff91d5a3fa0cda2b57d4abfc8f0bd44b6fb2725
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4526.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:d2fe258f6e40e256778227dfe586b0e1922e44281a37850beb3c96194d4ceb9b
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4527.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:3cf0b9d1633fdaec5f3843d97c6ff47415dddcf441ba5f3fda56d018fe098c71
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4528.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:e4aae50e191cdb663985b1a74486cf58deeed46310e24bb024f4542292a09dde
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4529.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:53413fdebae465f3c203374015948ab683da95e9a83940e562517b2cd7e86b91
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4530.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:4f9fd430e148d7bd95d53ac38119d5a9a160478fc63354c06600d3191d272319
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4531.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:d970b8fafc050ddbf72f7cb6ca8cc323f15975512e8544b4996ed392f015a58c
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4532.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c9a9c5601d32657b9171f636839b8e802568be3eca7b2bb2623ca24dd5e1d7fa
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4533.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:bac4f79fbdc8112e6cba025fb1e8bbc470a51507c9c206cc97282108d793c1f2
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4534.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:17c60a76aa0bac8276dbaa903aa343742a9e60c17d8e57a3fb705b31ede7ab7f
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4535.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:a1e1a166a4b1ad7565237fcea7b99b83240581cc767e11d90392323955014ee0
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4536.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:ab1d758618d7beb33ebedcc1d1f1c4fb21066c4939b0254722cadb32056265f0
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4537.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:65db382062b5802137f90b1755c827f0d236800e7bd1ef6d1e4fda9ddf390c2c
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4538.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:aca0980ea07f9dd61754d59bdd58747c2fc7cef7931bb8ce054b50323c256d5b
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4539.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:466e8c67ba809fe225dfc32e2630e682ec0ed9a1bdba8c46326eb48dc04c47a6
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4540.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c3e1ad050b0d33e52ddcba9e50845cb825a3e2b4119a15c7bba2ee079b76af61
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4541.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:6f0081a35a054ce3ef6f6948050f831a0b72954cd5682b3027e3996bf7d3edc1
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4542.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:55177ae00941e4635c4d0ee99257ee2ee61f38f8d6c0bea39d760cdad46b9296
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4543.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:ff6e44f50b9d130759a7c721e234ef2237fc8a9a67ef6e59d6420297478ce1eb
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4544.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:28f303e680e81a7966f1466f3eb160fe43ad4dc12924b39d69723cd5169ee16c
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4545.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:cd2d5dd6336d9be328859275c5bc11b231c35b187ae037addca9fd57ff322fe7
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4546.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:5815f940628edf4639f2dcc74c5be4ad88d48d24ea720b9649eb410b69c94fcd
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4547.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:9bd16ae4ad65391508216d80fced01ff787784f0cc54275b0181487f54c6156a
3
- size 2073864
 
 
 
 
data/semisupervised_set/depth_dpt/part0/atanasie_DJI_0652_full_4548.npz DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:23b17c3114c1d166645a778e86c7c9ccfeb6735b435d4dd383f2a56ed95866f2
3
- size 2073864