Hugging Face's logo

qualcomm
/
MaskRCNN

Image Segmentation
PyTorch
android
Model card Files
xet
 Community
qaihm-bot commited on
Commit
4a1e9d3
·
verified ·
1 Parent(s): 6fcd54f

v0.60.0

Browse files

See https://github.com/qualcomm/ai-hub-models/releases/v0.60.0 for changelog.

Files changed (2) hide show
  1. README.md +61 -40
  2. release_assets.json +9 -2
README.md CHANGED
@@ -14,7 +14,7 @@ pipeline_tag: image-segmentation
14
  Mask R-CNN is a machine learning model that extends Faster R-CNN to perform instance segmentation by detecting objects in an image while simultaneously generating a high-quality segmentation mask for each instance. It adds a branch for predicting segmentation masks in parallel with the existing branch for bounding box recognition.
15
 
16
  This is based on the implementation of MaskRCNN found [here](https://github.com/pytorch/vision).
17
- This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/maskrcnn) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
18
 
19
  Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
20
 
@@ -27,70 +27,91 @@ Below are pre-exported model assets ready for deployment.
27
 
28
  | Runtime | Precision | Chipset | SDK Versions | Download |
29
  |---|---|---|---|---|
30
- | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/releases/v0.59.0/maskrcnn-qnn_dlc-float.zip)
 
31
 
32
  For more device-specific assets and performance metrics, visit **[MaskRCNN on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/maskrcnn)**.
33
 
34
 
35
  ### Option 2: Export with Custom Configurations
36
 
37
- Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/maskrcnn) Python library to compile and export the model with your own:
38
  - Custom weights (e.g., fine-tuned checkpoints)
39
  - Custom input shapes
40
  - Target device and runtime configurations
41
 
42
  This option is ideal if you need to customize the model beyond the default configuration provided here.
43
 
44
- See our repository for [MaskRCNN on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/maskrcnn) for usage instructions.
45
 
46
  ## Model Details
47
 
48
  **Model Type:** Model_use_case.semantic_segmentation
49
 
50
  **Model Stats:**
51
- - Model checkpoint: Mask R-CNN ResNet-50 FPN V2
52
  - Input resolution: 800x800
 
 
53
  - Number of output classes: 91
54
  - Number of parameters: 46.4M
55
- - Model size (float): 177 MB
56
 
57
  ## Performance Summary
58
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
59
  |---|---|---|---|---|---|---
60
- | proposal_generator | QNN_DLC | float | Snapdragon® X2 Elite | 42.554 ms | 7 - 7 MB | NPU
61
- | proposal_generator | QNN_DLC | float | Snapdragon® X Elite | 91.369 ms | 7 - 7 MB | NPU
62
- | proposal_generator | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 70.212 ms | 7 - 2326 MB | NPU
63
- | proposal_generator | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 151.831 ms | 7 - 2736 MB | NPU
64
- | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 354.445 ms | 2 - 1759 MB | NPU
65
- | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 94.473 ms | 7 - 24 MB | NPU
66
- | proposal_generator | QNN_DLC | float | Qualcomm® SA8775P | 122.283 ms | 2 - 1759 MB | NPU
67
- | proposal_generator | QNN_DLC | float | Qualcomm® SA8650P | 122.283 ms | 2 - 1759 MB | NPU
68
- | proposal_generator | QNN_DLC | float | Qualcomm® SA8255P | 122.283 ms | 2 - 1759 MB | NPU
69
- | proposal_generator | QNN_DLC | float | Qualcomm® QCS8450 | 151.831 ms | 7 - 2736 MB | NPU
70
- | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 122.578 ms | 7 - 71 MB | NPU
71
- | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 91.369 ms | 7 - 7 MB | NPU
72
- | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 53.816 ms | 12 - 1523 MB | NPU
73
- | proposal_generator | QNN_DLC | float | Qualcomm® SA7255P | 354.445 ms | 2 - 1759 MB | NPU
74
- | proposal_generator | QNN_DLC | float | Qualcomm® SA8295P | 125.987 ms | 0 - 1430 MB | NPU
75
- | proposal_generator | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 53.816 ms | 12 - 1523 MB | NPU
76
- | proposal_generator | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 42.27 ms | 7 - 1789 MB | NPU
77
- | roi_head | QNN_DLC | float | Snapdragon® X2 Elite | 98.312 ms | 52 - 52 MB | NPU
78
- | roi_head | QNN_DLC | float | Snapdragon® X Elite | 257.611 ms | 52 - 52 MB | NPU
79
- | roi_head | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 177.662 ms | 49 - 897 MB | NPU
80
- | roi_head | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 300.661 ms | 0 - 918 MB | NPU
81
- | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 573.44 ms | 30 - 725 MB | NPU
82
- | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 243.427 ms | 52 - 679 MB | NPU
83
- | roi_head | QNN_DLC | float | Qualcomm® SA8775P | 270.285 ms | 49 - 925 MB | NPU
84
- | roi_head | QNN_DLC | float | Qualcomm® SA8650P | 270.285 ms | 49 - 925 MB | NPU
85
- | roi_head | QNN_DLC | float | Qualcomm® SA8255P | 270.285 ms | 49 - 925 MB | NPU
86
- | roi_head | QNN_DLC | float | Qualcomm® QCS8450 | 300.661 ms | 0 - 918 MB | NPU
87
- | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 279.963 ms | 52 - 106 MB | NPU
88
- | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 257.611 ms | 52 - 52 MB | NPU
89
- | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 125.217 ms | 34 - 726 MB | NPU
90
- | roi_head | QNN_DLC | float | Qualcomm® SA7255P | 573.44 ms | 30 - 725 MB | NPU
91
- | roi_head | QNN_DLC | float | Qualcomm® SA8295P | 306.965 ms | 49 - 847 MB | NPU
92
- | roi_head | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 125.217 ms | 34 - 726 MB | NPU
93
- | roi_head | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 92.646 ms | 13 - 719 MB | NPU
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
94
 
95
  ## License
96
  * The license for the original implementation of MaskRCNN can be found
 
14
  Mask R-CNN is a machine learning model that extends Faster R-CNN to perform instance segmentation by detecting objects in an image while simultaneously generating a high-quality segmentation mask for each instance. It adds a branch for predicting segmentation masks in parallel with the existing branch for bounding box recognition.
15
 
16
  This is based on the implementation of MaskRCNN found [here](https://github.com/pytorch/vision).
17
+ This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/maskrcnn) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
18
 
19
  Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
20
 
 
27
 
28
  | Runtime | Precision | Chipset | SDK Versions | Download |
29
  |---|---|---|---|---|
30
+ | ONNX | float | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/releases/v0.60.0/maskrcnn-onnx-float.zip)
31
+ | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/releases/v0.60.0/maskrcnn-qnn_dlc-float.zip)
32
 
33
  For more device-specific assets and performance metrics, visit **[MaskRCNN on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/maskrcnn)**.
34
 
35
 
36
  ### Option 2: Export with Custom Configurations
37
 
38
+ Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/maskrcnn) Python library to compile and export the model with your own:
39
  - Custom weights (e.g., fine-tuned checkpoints)
40
  - Custom input shapes
41
  - Target device and runtime configurations
42
 
43
  This option is ideal if you need to customize the model beyond the default configuration provided here.
44
 
45
+ See our repository for [MaskRCNN on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/maskrcnn) for usage instructions.
46
 
47
  ## Model Details
48
 
49
  **Model Type:** Model_use_case.semantic_segmentation
50
 
51
  **Model Stats:**
 
52
  - Input resolution: 800x800
53
+ - Model checkpoint: Mask R-CNN ResNet-50 FPN V2
54
+ - Model size (float): 177 MB
55
  - Number of output classes: 91
56
  - Number of parameters: 46.4M
 
57
 
58
  ## Performance Summary
59
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
60
  |---|---|---|---|---|---|---
61
+ | proposal_generator | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 68.787 ms | 20 - 2832 MB | NPU
62
+ | proposal_generator | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 161.492 ms | 138 - 2872 MB | NPU
63
+ | proposal_generator | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 125.168 ms | 7 - 18 MB | NPU
64
+ | proposal_generator | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 91.452 ms | 123 - 198 MB | NPU
65
+ | proposal_generator | ONNX | float | Qualcomm® QCS8450 | 161.492 ms | 138 - 2872 MB | NPU
66
+ | proposal_generator | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 118.666 ms | 7 - 17 MB | NPU
67
+ | proposal_generator | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 52.655 ms | 134 - 1631 MB | NPU
68
+ | proposal_generator | ONNX | float | Snapdragon® 8 Elite Mobile | 52.655 ms | 134 - 1631 MB | NPU
69
+ | proposal_generator | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 41.585 ms | 111 - 1464 MB | NPU
70
+ | proposal_generator | QNN_DLC | float | Snapdragon® X2 Elite | 42.783 ms | 7 - 7 MB | NPU
71
+ | proposal_generator | QNN_DLC | float | Snapdragon® X Elite | 91.275 ms | 7 - 7 MB | NPU
72
+ | proposal_generator | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 69.5 ms | 8 - 2329 MB | NPU
73
+ | proposal_generator | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 156.172 ms | 7 - 2736 MB | NPU
74
+ | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 127.502 ms | 7 - 72 MB | NPU
75
+ | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 354.018 ms | 2 - 1759 MB | NPU
76
+ | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 95.318 ms | 8 - 10 MB | NPU
77
+ | proposal_generator | QNN_DLC | float | Qualcomm® SA8775P | 122.279 ms | 0 - 1756 MB | NPU
78
+ | proposal_generator | QNN_DLC | float | Qualcomm® SA8650P | 122.279 ms | 0 - 1756 MB | NPU
79
+ | proposal_generator | QNN_DLC | float | Qualcomm® SA8255P | 122.279 ms | 0 - 1756 MB | NPU
80
+ | proposal_generator | QNN_DLC | float | Qualcomm® QCS8450 | 156.172 ms | 7 - 2736 MB | NPU
81
+ | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 122.416 ms | 7 - 71 MB | NPU
82
+ | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 91.275 ms | 7 - 7 MB | NPU
83
+ | proposal_generator | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 53.567 ms | 7 - 1517 MB | NPU
84
+ | proposal_generator | QNN_DLC | float | Qualcomm® SA7255P | 354.018 ms | 2 - 1759 MB | NPU
85
+ | proposal_generator | QNN_DLC | float | Qualcomm® SA8295P | 125.889 ms | 0 - 1429 MB | NPU
86
+ | proposal_generator | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 53.567 ms | 7 - 1517 MB | NPU
87
+ | proposal_generator | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 42.306 ms | 7 - 1788 MB | NPU
88
+ | roi_head | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 71.346 ms | 240 - 717 MB | NPU
89
+ | roi_head | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 139.492 ms | 233 - 679 MB | NPU
90
+ | roi_head | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 153.951 ms | 108 - 207 MB | NPU
91
+ | roi_head | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 90.01 ms | 271 - 274 MB | NPU
92
+ | roi_head | ONNX | float | Qualcomm® QCS8450 | 139.492 ms | 233 - 679 MB | NPU
93
+ | roi_head | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 113.168 ms | 112 - 211 MB | NPU
94
+ | roi_head | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 60.535 ms | 228 - 624 MB | NPU
95
+ | roi_head | ONNX | float | Snapdragon® 8 Elite Mobile | 60.535 ms | 228 - 624 MB | NPU
96
+ | roi_head | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 49.207 ms | 202 - 601 MB | NPU
97
+ | roi_head | QNN_DLC | float | Snapdragon® X2 Elite | 90.554 ms | 52 - 52 MB | NPU
98
+ | roi_head | QNN_DLC | float | Snapdragon® X Elite | 225.522 ms | 52 - 52 MB | NPU
99
+ | roi_head | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 160.521 ms | 22 - 808 MB | NPU
100
+ | roi_head | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 303.75 ms | 13 - 762 MB | NPU
101
+ | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 240.633 ms | 52 - 107 MB | NPU
102
+ | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 573.118 ms | 47 - 551 MB | NPU
103
+ | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 223.624 ms | 39 - 41 MB | NPU
104
+ | roi_head | QNN_DLC | float | Qualcomm® SA8775P | 259.583 ms | 39 - 768 MB | NPU
105
+ | roi_head | QNN_DLC | float | Qualcomm® SA8650P | 259.583 ms | 39 - 768 MB | NPU
106
+ | roi_head | QNN_DLC | float | Qualcomm® SA8255P | 259.583 ms | 39 - 768 MB | NPU
107
+ | roi_head | QNN_DLC | float | Qualcomm® QCS8450 | 303.75 ms | 13 - 762 MB | NPU
108
+ | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 255.693 ms | 52 - 106 MB | NPU
109
+ | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 225.522 ms | 52 - 52 MB | NPU
110
+ | roi_head | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 117.182 ms | 39 - 528 MB | NPU
111
+ | roi_head | QNN_DLC | float | Qualcomm® SA7255P | 573.118 ms | 47 - 551 MB | NPU
112
+ | roi_head | QNN_DLC | float | Qualcomm® SA8295P | 287.998 ms | 49 - 770 MB | NPU
113
+ | roi_head | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 117.182 ms | 39 - 528 MB | NPU
114
+ | roi_head | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 82.84 ms | 11 - 511 MB | NPU
115
 
116
  ## License
117
  * The license for the original implementation of MaskRCNN can be found
release_assets.json CHANGED
@@ -1,13 +1,20 @@
1
  {
2
- "version": "0.59.0",
3
  "precisions": {
4
  "float": {
5
  "universal_assets": {
 
 
 
 
 
 
 
6
  "qnn_dlc": {
7
  "tool_versions": {
8
  "qairt": "2.45.0.260326154327"
9
  },
10
- "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/releases/v0.59.0/maskrcnn-qnn_dlc-float.zip"
11
  }
12
  }
13
  }
 
1
  {
2
+ "version": "0.60.0",
3
  "precisions": {
4
  "float": {
5
  "universal_assets": {
6
+ "onnx": {
7
+ "tool_versions": {
8
+ "qairt": "2.45.0.260326154327",
9
+ "onnx_runtime": "1.27.1"
10
+ },
11
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/releases/v0.60.0/maskrcnn-onnx-float.zip"
12
+ },
13
  "qnn_dlc": {
14
  "tool_versions": {
15
  "qairt": "2.45.0.260326154327"
16
  },
17
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/releases/v0.60.0/maskrcnn-qnn_dlc-float.zip"
18
  }
19
  }
20
  }