Hugging Face's logo

qualcomm
/
MediaPipe-Hand-Detection

Object Detection
PyTorch
real_time
android
Model card Files
xet
Community
1
qaihm-bot commited on
Commit
6bba07a
·
verified ·
1 Parent(s): 18546b8

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +52 -28
README.md CHANGED
@@ -15,7 +15,7 @@ tags:
15
 
16
  The MediaPipe Hand Landmark Detector is a machine learning pipeline that predicts bounding boxes and pose skeletons of hands in an image.
17
 
18
- This model is an implementation of MediaPipe-Hand-Detection found [here](https://github.com/zmurez/MediaPipePyTorch/).
19
  This repository provides scripts to run MediaPipe-Hand-Detection on Qualcomm® devices.
20
  More details on model performance across various devices, can be found
21
  [here](https://aihub.qualcomm.com/models/mediapipe_hand).
@@ -31,17 +31,35 @@ More details on model performance across various devices, can be found
31
  - Number of parameters (MediaPipeHandLandmarkDetector): 2.01M
32
  - Model size (MediaPipeHandLandmarkDetector): 7.71 MB
33
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
34
 
35
 
36
 
37
- | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
38
- | ---|---|---|---|---|---|---|---|
39
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | TFLite | 0.714 ms | 0 - 5 MB | FP16 | NPU | [MediaPipeHandDetector.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite)
40
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | TFLite | 1.048 ms | 0 - 55 MB | FP16 | NPU | [MediaPipeHandLandmarkDetector.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite)
41
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | QNN Model Library | 0.791 ms | 1 - 20 MB | FP16 | NPU | [MediaPipeHandDetector.so](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.so)
42
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | QNN Model Library | 1.109 ms | 2 - 39 MB | FP16 | NPU | [MediaPipeHandLandmarkDetector.so](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.so)
43
-
44
-
45
 
46
  ## Installation
47
 
@@ -96,23 +114,25 @@ device. This script does the following:
96
  ```bash
97
  python -m qai_hub_models.models.mediapipe_hand.export
98
  ```
99
-
100
  ```
101
- Profile Job summary of MediaPipeHandDetector
102
- --------------------------------------------------
103
- Device: Snapdragon X Elite CRD (11)
104
- Estimated Inference Time: 0.93 ms
105
- Estimated Peak Memory Range: 0.75-0.75 MB
106
- Compute Units: NPU (195) | Total (195)
107
-
108
- Profile Job summary of MediaPipeHandLandmarkDetector
109
- --------------------------------------------------
110
- Device: Snapdragon X Elite CRD (11)
111
- Estimated Inference Time: 1.34 ms
112
- Estimated Peak Memory Range: 0.75-0.75 MB
113
- Compute Units: NPU (208) | Total (208)
114
-
115
-
 
 
 
116
  ```
117
 
118
 
@@ -240,15 +260,19 @@ provides instructions on how to use the `.so` shared library in an Android appl
240
  Get more details on MediaPipe-Hand-Detection's performance across various devices [here](https://aihub.qualcomm.com/models/mediapipe_hand).
241
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
242
 
 
243
  ## License
244
- - The license for the original implementation of MediaPipe-Hand-Detection can be found
245
- [here](https://github.com/zmurez/MediaPipePyTorch/blob/master/LICENSE).
246
- - 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)
 
247
 
248
  ## References
249
  * [MediaPipe Hands: On-device Real-time Hand Tracking](https://arxiv.org/abs/2006.10214)
250
  * [Source Model Implementation](https://github.com/zmurez/MediaPipePyTorch/)
251
 
 
 
252
  ## Community
253
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
254
  * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
 
15
 
16
  The MediaPipe Hand Landmark Detector is a machine learning pipeline that predicts bounding boxes and pose skeletons of hands in an image.
17
 
18
+ This model is an implementation of MediaPipe-Hand-Detection found [here]({source_repo}).
19
  This repository provides scripts to run MediaPipe-Hand-Detection on Qualcomm® devices.
20
  More details on model performance across various devices, can be found
21
  [here](https://aihub.qualcomm.com/models/mediapipe_hand).
 
31
  - Number of parameters (MediaPipeHandLandmarkDetector): 2.01M
32
  - Model size (MediaPipeHandLandmarkDetector): 7.71 MB
33
 
34
+ | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
35
+ |---|---|---|---|---|---|---|---|---|
36
+ | MediaPipeHandDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 0.704 ms | 0 - 4 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
37
+ | MediaPipeHandDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 1.16 ms | 0 - 17 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.onnx) |
38
+ | MediaPipeHandDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.612 ms | 0 - 59 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
39
+ | MediaPipeHandDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 0.903 ms | 0 - 67 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.onnx) |
40
+ | MediaPipeHandDetector | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 0.706 ms | 0 - 113 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
41
+ | MediaPipeHandDetector | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 0.711 ms | 0 - 61 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
42
+ | MediaPipeHandDetector | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 0.706 ms | 0 - 3 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
43
+ | MediaPipeHandDetector | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 0.708 ms | 0 - 3 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
44
+ | MediaPipeHandDetector | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 1.321 ms | 0 - 52 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
45
+ | MediaPipeHandDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.529 ms | 0 - 28 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.tflite) |
46
+ | MediaPipeHandDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 0.878 ms | 0 - 32 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.onnx) |
47
+ | MediaPipeHandDetector | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 1.204 ms | 6 - 6 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandDetector.onnx) |
48
+ | MediaPipeHandLandmarkDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 1.03 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
49
+ | MediaPipeHandLandmarkDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 1.552 ms | 0 - 8 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.onnx) |
50
+ | MediaPipeHandLandmarkDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.848 ms | 0 - 62 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
51
+ | MediaPipeHandLandmarkDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 1.213 ms | 0 - 65 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.onnx) |
52
+ | MediaPipeHandLandmarkDetector | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 1.003 ms | 0 - 171 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
53
+ | MediaPipeHandLandmarkDetector | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 1.008 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
54
+ | MediaPipeHandLandmarkDetector | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 1.004 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
55
+ | MediaPipeHandLandmarkDetector | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 1.035 ms | 0 - 1 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
56
+ | MediaPipeHandLandmarkDetector | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 2.59 ms | 0 - 55 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
57
+ | MediaPipeHandLandmarkDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.585 ms | 0 - 32 MB | FP16 | NPU | [MediaPipe-Hand-Detection.tflite](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.tflite) |
58
+ | MediaPipeHandLandmarkDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 1.068 ms | 0 - 37 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.onnx) |
59
+ | MediaPipeHandLandmarkDetector | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 1.641 ms | 8 - 8 MB | FP16 | NPU | [MediaPipe-Hand-Detection.onnx](https://huggingface.co/qualcomm/MediaPipe-Hand-Detection/blob/main/MediaPipeHandLandmarkDetector.onnx) |
60
 
61
 
62
 
 
 
 
 
 
 
 
 
63
 
64
  ## Installation
65
 
 
114
  ```bash
115
  python -m qai_hub_models.models.mediapipe_hand.export
116
  ```
 
117
  ```
118
+ Profiling Results
119
+ ------------------------------------------------------------
120
+ MediaPipeHandDetector
121
+ Device : Samsung Galaxy S23 (13)
122
+ Runtime : TFLITE
123
+ Estimated inference time (ms) : 0.7
124
+ Estimated peak memory usage (MB): [0, 4]
125
+ Total # Ops : 149
126
+ Compute Unit(s) : NPU (149 ops)
127
+
128
+ ------------------------------------------------------------
129
+ MediaPipeHandLandmarkDetector
130
+ Device : Samsung Galaxy S23 (13)
131
+ Runtime : TFLITE
132
+ Estimated inference time (ms) : 1.0
133
+ Estimated peak memory usage (MB): [0, 1]
134
+ Total # Ops : 158
135
+ Compute Unit(s) : NPU (158 ops)
136
  ```
137
 
138
 
 
260
  Get more details on MediaPipe-Hand-Detection's performance across various devices [here](https://aihub.qualcomm.com/models/mediapipe_hand).
261
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
262
 
263
+
264
  ## License
265
+ * The license for the original implementation of MediaPipe-Hand-Detection can be found [here](https://github.com/zmurez/MediaPipePyTorch/blob/master/LICENSE).
266
+ * 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)
267
+
268
+
269
 
270
  ## References
271
  * [MediaPipe Hands: On-device Real-time Hand Tracking](https://arxiv.org/abs/2006.10214)
272
  * [Source Model Implementation](https://github.com/zmurez/MediaPipePyTorch/)
273
 
274
+
275
+
276
  ## Community
277
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
278
  * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).