Hugging Face's logo

qualcomm
/
HRNetFace

Object Detection
PyTorch
LiteRT
real_time
android
Model card Files
xet
Community
qaihm-bot commited on
Commit
ba41173
·
verified ·
1 Parent(s): ab6ff92

v0.41.2

Browse files

See https://github.com/quic/ai-hub-models/releases/v0.41.2 for changelog.

Files changed (11) hide show
  1. .gitattributes +3 -0
  2. DEPLOYMENT_MODEL_LICENSE.pdf +3 -0
  3. HRNetFace_float.dlc +3 -0
  4. HRNetFace_float.onnx.zip +3 -0
  5. HRNetFace_float.tflite +3 -0
  6. HRNetFace_w8a8.dlc +3 -0
  7. HRNetFace_w8a8.onnx.zip +3 -0
  8. HRNetFace_w8a8.tflite +3 -0
  9. LICENSE +2 -0
  10. README.md +294 -0
  11. tool-versions.yaml +4 -0
.gitattributes CHANGED
@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ DEPLOYMENT_MODEL_LICENSE.pdf filter=lfs diff=lfs merge=lfs -text
37
+ HRNetFace_float.dlc filter=lfs diff=lfs merge=lfs -text
38
+ HRNetFace_w8a8.dlc filter=lfs diff=lfs merge=lfs -text
DEPLOYMENT_MODEL_LICENSE.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4409f93b0e82531303b3e10f52f1fdfb56467a25f05b7441c6bbd8bb8a64b42c
3
+ size 109629
HRNetFace_float.dlc ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f31c868a21e34f3967935f61bec9f68414e7a102c1a9991ce1a63a084344a12
3
+ size 39287364
HRNetFace_float.onnx.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:efc1943448009fc6453802f947c49e69ca5b95112c8057182ee87ceb8a24377e
3
+ size 35984397
HRNetFace_float.tflite ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94e2d5a52ee5b54cec77f260e70d7419b50707ab2520066b820b8f95e4653bf8
3
+ size 38672576
HRNetFace_w8a8.dlc ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1e61517ac94abc0c879270a8ea2da18169766aa9edb9f27d250079679f0f308
3
+ size 11472044
HRNetFace_w8a8.onnx.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1bf357ed85379381bb677ad3cbe8cc707c07d0a880e8a87062cfda9805639368
3
+ size 18582273
HRNetFace_w8a8.tflite ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:202a2d36f2a6792e382294bb4cb1b2287dfe6dea7b3c736fece47c468eaf2965
3
+ size 10331344
LICENSE ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ The license of the original trained model can be found at https://github.com/HRNet/HRNet-Facial-Landmark-Detection/blob/master/LICENCE.
2
+ The license for the deployable model files (.tflite, .onnx, .dlc, .bin, etc.) can be found in DEPLOYMENT_MODEL_LICENSE.pdf.
README.md ADDED
@@ -0,0 +1,294 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ license: other
4
+ tags:
5
+ - real_time
6
+ - android
7
+ pipeline_tag: object-detection
8
+
9
+ ---
10
+
11
+ ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_face/web-assets/model_demo.png)
12
+
13
+ # HRNetFace: Optimized for Mobile Deployment
14
+ ## Comprehensive facial analysis by extracting face features
15
+
16
+
17
+ Detects attributes (liveness, eye closeness, mask presence, glasses presence, sunglasses presence) that apply to a given face.
18
+
19
+ This model is an implementation of HRNetFace found [here](https://github.com/HRNet/HRNet-Facial-Landmark-Detection).
20
+
21
+
22
+ This repository provides scripts to run HRNetFace on Qualcomm® devices.
23
+ More details on model performance across various devices, can be found
24
+ [here](https://aihub.qualcomm.com/models/hrnet_face).
25
+
26
+
27
+
28
+ ### Model Details
29
+
30
+ - **Model Type:** Model_use_case.object_detection
31
+ - **Model Stats:**
32
+ - Model checkpoint: HR18-COFW.pth
33
+ - Input resolution: 256x256
34
+ - Number of parameters: 9.68M
35
+ - Model size (float): 36.87MB
36
+ - Model size (w8a8): 17.7 MB
37
+
38
+ | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
39
+ |---|---|---|---|---|---|---|---|---|
40
+ | HRNetFace | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 15.596 ms | 0 - 58 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
41
+ | HRNetFace | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 15.524 ms | 1 - 44 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
42
+ | HRNetFace | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 4.473 ms | 0 - 61 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
43
+ | HRNetFace | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 4.812 ms | 1 - 46 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
44
+ | HRNetFace | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 3.142 ms | 0 - 130 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
45
+ | HRNetFace | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 3.231 ms | 1 - 14 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
46
+ | HRNetFace | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 3.287 ms | 0 - 46 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) |
47
+ | HRNetFace | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 21.535 ms | 0 - 59 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
48
+ | HRNetFace | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 5.13 ms | 1 - 45 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
49
+ | HRNetFace | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 15.596 ms | 0 - 58 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
50
+ | HRNetFace | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 15.524 ms | 1 - 44 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
51
+ | HRNetFace | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 3.15 ms | 0 - 130 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
52
+ | HRNetFace | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 3.233 ms | 1 - 15 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
53
+ | HRNetFace | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 5.515 ms | 0 - 53 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
54
+ | HRNetFace | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 5.567 ms | 1 - 44 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
55
+ | HRNetFace | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 3.148 ms | 0 - 132 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
56
+ | HRNetFace | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 3.24 ms | 2 - 16 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
57
+ | HRNetFace | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 21.535 ms | 0 - 59 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
58
+ | HRNetFace | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 5.13 ms | 1 - 45 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
59
+ | HRNetFace | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 2.217 ms | 0 - 66 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
60
+ | HRNetFace | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 2.323 ms | 0 - 53 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
61
+ | HRNetFace | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 2.34 ms | 0 - 68 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) |
62
+ | HRNetFace | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 1.737 ms | 0 - 61 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
63
+ | HRNetFace | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 1.77 ms | 1 - 49 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
64
+ | HRNetFace | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 1.846 ms | 0 - 56 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) |
65
+ | HRNetFace | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 1.419 ms | 0 - 59 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) |
66
+ | HRNetFace | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 1.406 ms | 1 - 50 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
67
+ | HRNetFace | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 1.536 ms | 0 - 58 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) |
68
+ | HRNetFace | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 3.625 ms | 28 - 28 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) |
69
+ | HRNetFace | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 3.222 ms | 30 - 30 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) |
70
+ | HRNetFace | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 3.246 ms | 0 - 44 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
71
+ | HRNetFace | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 3.342 ms | 0 - 44 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
72
+ | HRNetFace | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 1.455 ms | 0 - 65 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
73
+ | HRNetFace | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 1.78 ms | 0 - 56 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
74
+ | HRNetFace | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 1.265 ms | 0 - 45 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
75
+ | HRNetFace | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 1.364 ms | 0 - 17 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
76
+ | HRNetFace | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 2.039 ms | 0 - 20 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) |
77
+ | HRNetFace | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 1.722 ms | 0 - 45 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
78
+ | HRNetFace | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 1.763 ms | 0 - 44 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
79
+ | HRNetFace | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 3.818 ms | 0 - 59 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
80
+ | HRNetFace | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 58.295 ms | 18 - 41 MB | CPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) |
81
+ | HRNetFace | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 18.609 ms | 0 - 3 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
82
+ | HRNetFace | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 48.157 ms | 16 - 30 MB | CPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) |
83
+ | HRNetFace | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 3.246 ms | 0 - 44 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
84
+ | HRNetFace | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 3.342 ms | 0 - 44 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
85
+ | HRNetFace | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 1.263 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
86
+ | HRNetFace | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 1.38 ms | 0 - 17 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
87
+ | HRNetFace | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 2.153 ms | 0 - 52 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
88
+ | HRNetFace | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 2.264 ms | 0 - 51 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
89
+ | HRNetFace | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 1.265 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
90
+ | HRNetFace | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 1.376 ms | 0 - 15 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
91
+ | HRNetFace | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 1.722 ms | 0 - 45 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
92
+ | HRNetFace | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 1.763 ms | 0 - 44 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
93
+ | HRNetFace | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.871 ms | 0 - 67 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
94
+ | HRNetFace | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.934 ms | 0 - 58 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
95
+ | HRNetFace | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 1.277 ms | 0 - 73 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) |
96
+ | HRNetFace | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.712 ms | 0 - 50 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
97
+ | HRNetFace | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.671 ms | 0 - 51 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
98
+ | HRNetFace | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 1.022 ms | 0 - 59 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) |
99
+ | HRNetFace | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.605 ms | 0 - 51 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) |
100
+ | HRNetFace | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.552 ms | 0 - 52 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
101
+ | HRNetFace | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.885 ms | 0 - 61 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) |
102
+ | HRNetFace | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 1.601 ms | 36 - 36 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) |
103
+ | HRNetFace | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 1.86 ms | 15 - 15 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) |
104
+
105
+
106
+
107
+
108
+ ## Installation
109
+
110
+
111
+ Install the package via pip:
112
+ ```bash
113
+ # NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
114
+ pip install "qai-hub-models[hrnet-face]"
115
+ ```
116
+
117
+
118
+ ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device
119
+
120
+ Sign-in to [Qualcomm® AI Hub](https://app.aihub.qualcomm.com/) with your
121
+ Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`.
122
+
123
+ With this API token, you can configure your client to run models on the cloud
124
+ hosted devices.
125
+ ```bash
126
+ qai-hub configure --api_token API_TOKEN
127
+ ```
128
+ Navigate to [docs](https://app.aihub.qualcomm.com/docs/) for more information.
129
+
130
+
131
+
132
+ ## Demo off target
133
+
134
+ The package contains a simple end-to-end demo that downloads pre-trained
135
+ weights and runs this model on a sample input.
136
+
137
+ ```bash
138
+ python -m qai_hub_models.models.hrnet_face.demo
139
+ ```
140
+
141
+ The above demo runs a reference implementation of pre-processing, model
142
+ inference, and post processing.
143
+
144
+ **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
145
+ environment, please add the following to your cell (instead of the above).
146
+ ```
147
+ %run -m qai_hub_models.models.hrnet_face.demo
148
+ ```
149
+
150
+
151
+ ### Run model on a cloud-hosted device
152
+
153
+ In addition to the demo, you can also run the model on a cloud-hosted Qualcomm®
154
+ device. This script does the following:
155
+ * Performance check on-device on a cloud-hosted device
156
+ * Downloads compiled assets that can be deployed on-device for Android.
157
+ * Accuracy check between PyTorch and on-device outputs.
158
+
159
+ ```bash
160
+ python -m qai_hub_models.models.hrnet_face.export
161
+ ```
162
+
163
+
164
+
165
+ ## How does this work?
166
+
167
+ This [export script](https://aihub.qualcomm.com/models/hrnet_face/qai_hub_models/models/HRNetFace/export.py)
168
+ leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model
169
+ on-device. Lets go through each step below in detail:
170
+
171
+ Step 1: **Compile model for on-device deployment**
172
+
173
+ To compile a PyTorch model for on-device deployment, we first trace the model
174
+ in memory using the `jit.trace` and then call the `submit_compile_job` API.
175
+
176
+ ```python
177
+ import torch
178
+
179
+ import qai_hub as hub
180
+ from qai_hub_models.models.hrnet_face import Model
181
+
182
+ # Load the model
183
+ torch_model = Model.from_pretrained()
184
+
185
+ # Device
186
+ device = hub.Device("Samsung Galaxy S25")
187
+
188
+ # Trace model
189
+ input_shape = torch_model.get_input_spec()
190
+ sample_inputs = torch_model.sample_inputs()
191
+
192
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
193
+
194
+ # Compile model on a specific device
195
+ compile_job = hub.submit_compile_job(
196
+ model=pt_model,
197
+ device=device,
198
+ input_specs=torch_model.get_input_spec(),
199
+ )
200
+
201
+ # Get target model to run on-device
202
+ target_model = compile_job.get_target_model()
203
+
204
+ ```
205
+
206
+
207
+ Step 2: **Performance profiling on cloud-hosted device**
208
+
209
+ After compiling models from step 1. Models can be profiled model on-device using the
210
+ `target_model`. Note that this scripts runs the model on a device automatically
211
+ provisioned in the cloud. Once the job is submitted, you can navigate to a
212
+ provided job URL to view a variety of on-device performance metrics.
213
+ ```python
214
+ profile_job = hub.submit_profile_job(
215
+ model=target_model,
216
+ device=device,
217
+ )
218
+
219
+ ```
220
+
221
+ Step 3: **Verify on-device accuracy**
222
+
223
+ To verify the accuracy of the model on-device, you can run on-device inference
224
+ on sample input data on the same cloud hosted device.
225
+ ```python
226
+ input_data = torch_model.sample_inputs()
227
+ inference_job = hub.submit_inference_job(
228
+ model=target_model,
229
+ device=device,
230
+ inputs=input_data,
231
+ )
232
+ on_device_output = inference_job.download_output_data()
233
+
234
+ ```
235
+ With the output of the model, you can compute like PSNR, relative errors or
236
+ spot check the output with expected output.
237
+
238
+ **Note**: This on-device profiling and inference requires access to Qualcomm®
239
+ AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup).
240
+
241
+
242
+
243
+ ## Run demo on a cloud-hosted device
244
+
245
+ You can also run the demo on-device.
246
+
247
+ ```bash
248
+ python -m qai_hub_models.models.hrnet_face.demo --eval-mode on-device
249
+ ```
250
+
251
+ **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
252
+ environment, please add the following to your cell (instead of the above).
253
+ ```
254
+ %run -m qai_hub_models.models.hrnet_face.demo -- --eval-mode on-device
255
+ ```
256
+
257
+
258
+ ## Deploying compiled model to Android
259
+
260
+
261
+ The models can be deployed using multiple runtimes:
262
+ - TensorFlow Lite (`.tflite` export): [This
263
+ tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a
264
+ guide to deploy the .tflite model in an Android application.
265
+
266
+
267
+ - QNN (`.so` export ): This [sample
268
+ app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html)
269
+ provides instructions on how to use the `.so` shared library in an Android application.
270
+
271
+
272
+ ## View on Qualcomm® AI Hub
273
+ Get more details on HRNetFace's performance across various devices [here](https://aihub.qualcomm.com/models/hrnet_face).
274
+ Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
275
+
276
+
277
+ ## License
278
+ * The license for the original implementation of HRNetFace can be found
279
+ [here](https://github.com/HRNet/HRNet-Facial-Landmark-Detection/blob/master/LICENCE).
280
+ * 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)
281
+
282
+
283
+
284
+ ## References
285
+ * [Deep High-Resolution Representation Learning for Visual Recognition](https://arxiv.org/abs/1908.07919)
286
+ * [Source Model Implementation](https://github.com/HRNet/HRNet-Facial-Landmark-Detection)
287
+
288
+
289
+
290
+ ## Community
291
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
292
+ * For questions or feedback please [reach out to us](mailto:[email protected]).
293
+
294
+
tool-versions.yaml ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ tool_versions:
2
+ onnx:
3
+ qairt: 2.37.1.250807093845_124904
4
+ onnx_runtime: 1.23.0