Hugging Face's logo

qualcomm
/
MediaPipe-Face-Detection-Quantized

Object Detection
PyTorch
TF Lite
ONNX
real_time
quantized
android
Model card Files Community
qaihm-bot commited on
Commit
cf98cbf
·
verified ·
1 Parent(s): 1d2ee53

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +68 -28
README.md CHANGED
@@ -16,7 +16,7 @@ tags:
16
 
17
  Designed for sub-millisecond processing, this model predicts bounding boxes and pose skeletons (left eye, right eye, nose tip, mouth, left eye tragion, and right eye tragion) of faces in an image.
18
 
19
- This model is an implementation of MediaPipe-Face-Detection-Quantized found [here](https://github.com/zmurez/MediaPipePyTorch/).
20
  This repository provides scripts to run MediaPipe-Face-Detection-Quantized on Qualcomm® devices.
21
  More details on model performance across various devices, can be found
22
  [here](https://aihub.qualcomm.com/models/mediapipe_face_quantized).
@@ -33,17 +33,51 @@ More details on model performance across various devices, can be found
33
  - Number of parameters (MediaPipeFaceLandmarkDetector): 603K
34
  - Model size (MediaPipeFaceLandmarkDetector): 746 KB
35
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
 
37
 
38
 
39
- | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
40
- | ---|---|---|---|---|---|---|---|
41
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | TFLite | 0.274 ms | 0 - 1 MB | FP16 | NPU | [MediaPipeFaceDetector.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite)
42
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | TFLite | 0.184 ms | 0 - 16 MB | FP16 | NPU | [MediaPipeFaceLandmarkDetector.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite)
43
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | QNN Model Library | 0.3 ms | 0 - 5 MB | FP16 | NPU | [MediaPipeFaceDetector.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.so)
44
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | QNN Model Library | 0.219 ms | 0 - 3 MB | FP16 | NPU | [MediaPipeFaceLandmarkDetector.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.so)
45
-
46
-
47
 
48
  ## Installation
49
 
@@ -98,23 +132,25 @@ device. This script does the following:
98
  ```bash
99
  python -m qai_hub_models.models.mediapipe_face_quantized.export
100
  ```
101
-
102
  ```
103
- Profile Job summary of MediaPipeFaceDetector
104
- --------------------------------------------------
105
- Device: Snapdragon X Elite CRD (11)
106
- Estimated Inference Time: 0.43 ms
107
- Estimated Peak Memory Range: 0.45-0.45 MB
108
- Compute Units: NPU (118) | Total (118)
109
-
110
- Profile Job summary of MediaPipeFaceLandmarkDetector
111
- --------------------------------------------------
112
- Device: Snapdragon X Elite CRD (11)
113
- Estimated Inference Time: 0.33 ms
114
- Estimated Peak Memory Range: 0.56-0.56 MB
115
- Compute Units: NPU (112) | Total (112)
116
-
117
-
 
 
 
118
  ```
119
 
120
 
@@ -139,15 +175,19 @@ provides instructions on how to use the `.so` shared library in an Android appl
139
  Get more details on MediaPipe-Face-Detection-Quantized's performance across various devices [here](https://aihub.qualcomm.com/models/mediapipe_face_quantized).
140
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
141
 
 
142
  ## License
143
- - The license for the original implementation of MediaPipe-Face-Detection-Quantized can be found
144
- [here](https://github.com/zmurez/MediaPipePyTorch/blob/master/LICENSE).
145
- - The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
 
146
 
147
  ## References
148
  * [BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs](https://arxiv.org/abs/1907.05047)
149
  * [Source Model Implementation](https://github.com/zmurez/MediaPipePyTorch/)
150
 
 
 
151
  ## Community
152
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
153
  * For questions or feedback please [reach out to us](mailto:[email protected]).
 
16
 
17
  Designed for sub-millisecond processing, this model predicts bounding boxes and pose skeletons (left eye, right eye, nose tip, mouth, left eye tragion, and right eye tragion) of faces in an image.
18
 
19
+ This model is an implementation of MediaPipe-Face-Detection-Quantized found [here]({source_repo}).
20
  This repository provides scripts to run MediaPipe-Face-Detection-Quantized on Qualcomm® devices.
21
  More details on model performance across various devices, can be found
22
  [here](https://aihub.qualcomm.com/models/mediapipe_face_quantized).
 
33
  - Number of parameters (MediaPipeFaceLandmarkDetector): 603K
34
  - Model size (MediaPipeFaceLandmarkDetector): 746 KB
35
 
36
+ | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
37
+ |---|---|---|---|---|---|---|---|---|
38
+ | MediaPipeFaceDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 0.275 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
39
+ | MediaPipeFaceDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 0.297 ms | 0 - 73 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.so) |
40
+ | MediaPipeFaceDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.182 ms | 0 - 32 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
41
+ | MediaPipeFaceDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 0.236 ms | 0 - 19 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.so) |
42
+ | MediaPipeFaceDetector | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 0.681 ms | 0 - 25 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
43
+ | MediaPipeFaceDetector | RB3 Gen 2 (Proxy) | QCS6490 Proxy | QNN | 0.762 ms | 0 - 8 MB | FP16 | NPU | Use Export Script |
44
+ | MediaPipeFaceDetector | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 5.031 ms | 0 - 5 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
45
+ | MediaPipeFaceDetector | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 0.273 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
46
+ | MediaPipeFaceDetector | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 0.301 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
47
+ | MediaPipeFaceDetector | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 0.279 ms | 0 - 70 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
48
+ | MediaPipeFaceDetector | SA8255 (Proxy) | SA8255P Proxy | QNN | 0.303 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
49
+ | MediaPipeFaceDetector | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 0.274 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
50
+ | MediaPipeFaceDetector | SA8775 (Proxy) | SA8775P Proxy | QNN | 0.303 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
51
+ | MediaPipeFaceDetector | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 0.272 ms | 0 - 2 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
52
+ | MediaPipeFaceDetector | SA8650 (Proxy) | SA8650P Proxy | QNN | 0.301 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
53
+ | MediaPipeFaceDetector | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 0.324 ms | 0 - 33 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
54
+ | MediaPipeFaceDetector | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 0.35 ms | 0 - 19 MB | FP16 | NPU | Use Export Script |
55
+ | MediaPipeFaceDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.202 ms | 0 - 23 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceDetector.tflite) |
56
+ | MediaPipeFaceDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 0.207 ms | 0 - 15 MB | FP16 | NPU | Use Export Script |
57
+ | MediaPipeFaceDetector | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 0.427 ms | 1 - 1 MB | FP16 | NPU | Use Export Script |
58
+ | MediaPipeFaceLandmarkDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 0.18 ms | 0 - 68 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
59
+ | MediaPipeFaceLandmarkDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 0.226 ms | 0 - 3 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.so) |
60
+ | MediaPipeFaceLandmarkDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.142 ms | 0 - 26 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
61
+ | MediaPipeFaceLandmarkDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 0.166 ms | 0 - 12 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.so) |
62
+ | MediaPipeFaceLandmarkDetector | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 0.395 ms | 0 - 19 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
63
+ | MediaPipeFaceLandmarkDetector | RB3 Gen 2 (Proxy) | QCS6490 Proxy | QNN | 0.49 ms | 0 - 8 MB | FP16 | NPU | Use Export Script |
64
+ | MediaPipeFaceLandmarkDetector | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 2.921 ms | 0 - 7 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
65
+ | MediaPipeFaceLandmarkDetector | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 0.182 ms | 0 - 3 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
66
+ | MediaPipeFaceLandmarkDetector | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 0.212 ms | 0 - 2 MB | FP16 | NPU | Use Export Script |
67
+ | MediaPipeFaceLandmarkDetector | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 0.185 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
68
+ | MediaPipeFaceLandmarkDetector | SA8255 (Proxy) | SA8255P Proxy | QNN | 0.216 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
69
+ | MediaPipeFaceLandmarkDetector | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 0.185 ms | 0 - 3 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
70
+ | MediaPipeFaceLandmarkDetector | SA8775 (Proxy) | SA8775P Proxy | QNN | 0.214 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
71
+ | MediaPipeFaceLandmarkDetector | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 0.182 ms | 0 - 2 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
72
+ | MediaPipeFaceLandmarkDetector | SA8650 (Proxy) | SA8650P Proxy | QNN | 0.219 ms | 0 - 1 MB | FP16 | NPU | Use Export Script |
73
+ | MediaPipeFaceLandmarkDetector | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 0.214 ms | 0 - 28 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
74
+ | MediaPipeFaceLandmarkDetector | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 0.269 ms | 0 - 14 MB | FP16 | NPU | Use Export Script |
75
+ | MediaPipeFaceLandmarkDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.12 ms | 0 - 18 MB | FP16 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/MediaPipeFaceLandmarkDetector.tflite) |
76
+ | MediaPipeFaceLandmarkDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 0.177 ms | 0 - 10 MB | FP16 | NPU | Use Export Script |
77
+ | MediaPipeFaceLandmarkDetector | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 0.343 ms | 1 - 1 MB | FP16 | NPU | Use Export Script |
78
 
79
 
80
 
 
 
 
 
 
 
 
 
81
 
82
  ## Installation
83
 
 
132
  ```bash
133
  python -m qai_hub_models.models.mediapipe_face_quantized.export
134
  ```
 
135
  ```
136
+ Profiling Results
137
+ ------------------------------------------------------------
138
+ MediaPipeFaceDetector
139
+ Device : Samsung Galaxy S23 (13)
140
+ Runtime : TFLITE
141
+ Estimated inference time (ms) : 0.3
142
+ Estimated peak memory usage (MB): [0, 1]
143
+ Total # Ops : 121
144
+ Compute Unit(s) : NPU (121 ops)
145
+
146
+ ------------------------------------------------------------
147
+ MediaPipeFaceLandmarkDetector
148
+ Device : Samsung Galaxy S23 (13)
149
+ Runtime : TFLITE
150
+ Estimated inference time (ms) : 0.2
151
+ Estimated peak memory usage (MB): [0, 68]
152
+ Total # Ops : 117
153
+ Compute Unit(s) : NPU (117 ops)
154
  ```
155
 
156
 
 
175
  Get more details on MediaPipe-Face-Detection-Quantized's performance across various devices [here](https://aihub.qualcomm.com/models/mediapipe_face_quantized).
176
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
177
 
178
+
179
  ## License
180
+ * The license for the original implementation of MediaPipe-Face-Detection-Quantized can be found [here](https://github.com/zmurez/MediaPipePyTorch/blob/master/LICENSE).
181
+ * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
182
+
183
+
184
 
185
  ## References
186
  * [BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs](https://arxiv.org/abs/1907.05047)
187
  * [Source Model Implementation](https://github.com/zmurez/MediaPipePyTorch/)
188
 
189
+
190
+
191
  ## Community
192
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
193
  * For questions or feedback please [reach out to us](mailto:[email protected]).