Hugging Face's logo

qualcomm
/
BiseNet

Other
PyTorch
deprecated
Model card Files
xet
 Community
qaihm-bot commited on
Commit
bef8d56
·
verified ·
1 Parent(s): 7ab2b6d

v0.49.1

Browse files

Deprecation notice.

Files changed (2) hide show
  1. LICENSE +0 -1
  2. README.md +3 -138
LICENSE DELETED
@@ -1 +0,0 @@
1
- The license of the original trained model can be found at https://github.com/ooooverflow/BiSeNet/pull/45/files.
 
 
README.md CHANGED
@@ -2,143 +2,8 @@
2
  library_name: pytorch
3
  license: other
4
  tags:
5
- - real_time
6
- - android
7
- pipeline_tag: image-segmentation
8
 
9
  ---
10
-
11
- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/bisenet/web-assets/model_demo.png)
12
-
13
- # BiseNet: Optimized for Qualcomm Devices
14
-
15
- BiSeNet (Bilateral Segmentation Network) is a novel architecture designed for real-time semantic segmentation. It addresses the challenge of balancing spatial resolution and receptive field by employing a Spatial Path to preserve high-resolution features and a context path to capture sufficient receptive field.
16
-
17
- This is based on the implementation of BiseNet found [here](https://github.com/ooooverflow/BiSeNet).
18
- 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/main/qai_hub_models/models/bisenet) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
19
-
20
- 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.
21
-
22
- ## Getting Started
23
- There are two ways to deploy this model on your device:
24
-
25
- ### Option 1: Download Pre-Exported Models
26
-
27
- Below are pre-exported model assets ready for deployment.
28
-
29
- | Runtime | Precision | Chipset | SDK Versions | Download |
30
- |---|---|---|---|---|
31
- | ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/bisenet/releases/v0.48.0/bisenet-onnx-float.zip)
32
- | ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/bisenet/releases/v0.48.0/bisenet-onnx-w8a8.zip)
33
- | QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/bisenet/releases/v0.48.0/bisenet-qnn_dlc-float.zip)
34
- | QNN_DLC | w8a8 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/bisenet/releases/v0.48.0/bisenet-qnn_dlc-w8a8.zip)
35
- | TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/bisenet/releases/v0.48.0/bisenet-tflite-float.zip)
36
- | TFLITE | w8a8 | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/bisenet/releases/v0.48.0/bisenet-tflite-w8a8.zip)
37
-
38
- For more device-specific assets and performance metrics, visit **[BiseNet on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/bisenet)**.
39
-
40
-
41
- ### Option 2: Export with Custom Configurations
42
-
43
- Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/bisenet) Python library to compile and export the model with your own:
44
- - Custom weights (e.g., fine-tuned checkpoints)
45
- - Custom input shapes
46
- - Target device and runtime configurations
47
-
48
- This option is ideal if you need to customize the model beyond the default configuration provided here.
49
-
50
- See our repository for [BiseNet on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/bisenet) for usage instructions.
51
-
52
- ## Model Details
53
-
54
- **Model Type:** Model_use_case.semantic_segmentation
55
-
56
- **Model Stats:**
57
- - Model checkpoint: best_dice_loss_miou_0.655.pth
58
- - Inference latency: RealTime
59
- - Input resolution: 720x960
60
- - Number of parameters: 12.0M
61
- - Model size (float): 45.7 MB
62
-
63
- ## Performance Summary
64
- | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
65
- |---|---|---|---|---|---|---
66
- | BiseNet | ONNX | float | Snapdragon® X2 Elite | 15.557 ms | 65 - 65 MB | NPU
67
- | BiseNet | ONNX | float | Snapdragon® X Elite | 30.818 ms | 66 - 66 MB | NPU
68
- | BiseNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 24.579 ms | 73 - 321 MB | NPU
69
- | BiseNet | ONNX | float | Qualcomm® QCS8550 (Proxy) | 32.09 ms | 0 - 2 MB | NPU
70
- | BiseNet | ONNX | float | Qualcomm® QCS9075 | 49.248 ms | 8 - 11 MB | NPU
71
- | BiseNet | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.118 ms | 66 - 255 MB | NPU
72
- | BiseNet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.059 ms | 62 - 262 MB | NPU
73
- | BiseNet | ONNX | w8a8 | Snapdragon® X2 Elite | 3.792 ms | 17 - 17 MB | NPU
74
- | BiseNet | ONNX | w8a8 | Snapdragon® X Elite | 8.676 ms | 19 - 19 MB | NPU
75
- | BiseNet | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 6.019 ms | 18 - 259 MB | NPU
76
- | BiseNet | ONNX | w8a8 | Qualcomm® QCS6490 | 235.121 ms | 221 - 235 MB | CPU
77
- | BiseNet | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 8.336 ms | 18 - 21 MB | NPU
78
- | BiseNet | ONNX | w8a8 | Qualcomm® QCS9075 | 10.171 ms | 18 - 21 MB | NPU
79
- | BiseNet | ONNX | w8a8 | Qualcomm® QCM6690 | 229.889 ms | 223 - 231 MB | CPU
80
- | BiseNet | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 4.735 ms | 17 - 215 MB | NPU
81
- | BiseNet | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 219.762 ms | 211 - 218 MB | CPU
82
- | BiseNet | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 3.808 ms | 18 - 219 MB | NPU
83
- | BiseNet | QNN_DLC | float | Snapdragon® X2 Elite | 14.465 ms | 8 - 8 MB | NPU
84
- | BiseNet | QNN_DLC | float | Snapdragon® X Elite | 28.599 ms | 8 - 8 MB | NPU
85
- | BiseNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 20.276 ms | 6 - 281 MB | NPU
86
- | BiseNet | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 107.409 ms | 2 - 191 MB | NPU
87
- | BiseNet | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 28.289 ms | 8 - 10 MB | NPU
88
- | BiseNet | QNN_DLC | float | Qualcomm® SA8775P | 38.884 ms | 1 - 188 MB | NPU
89
- | BiseNet | QNN_DLC | float | Qualcomm® QCS9075 | 56.203 ms | 8 - 49 MB | NPU
90
- | BiseNet | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 61.868 ms | 8 - 279 MB | NPU
91
- | BiseNet | QNN_DLC | float | Qualcomm® SA7255P | 107.409 ms | 2 - 191 MB | NPU
92
- | BiseNet | QNN_DLC | float | Qualcomm® SA8295P | 44.168 ms | 0 - 213 MB | NPU
93
- | BiseNet | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 15.366 ms | 0 - 219 MB | NPU
94
- | BiseNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 12.754 ms | 8 - 287 MB | NPU
95
- | BiseNet | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 4.842 ms | 2 - 2 MB | NPU
96
- | BiseNet | QNN_DLC | w8a8 | Snapdragon® X Elite | 10.09 ms | 2 - 2 MB | NPU
97
- | BiseNet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 6.681 ms | 2 - 231 MB | NPU
98
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 40.298 ms | 2 - 13 MB | NPU
99
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 19.98 ms | 2 - 182 MB | NPU
100
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 9.439 ms | 2 - 170 MB | NPU
101
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® SA8775P | 10.279 ms | 2 - 183 MB | NPU
102
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 13.412 ms | 1 - 12 MB | NPU
103
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 89.681 ms | 2 - 206 MB | NPU
104
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 16.162 ms | 2 - 231 MB | NPU
105
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® SA7255P | 19.98 ms | 2 - 182 MB | NPU
106
- | BiseNet | QNN_DLC | w8a8 | Qualcomm® SA8295P | 12.693 ms | 2 - 185 MB | NPU
107
- | BiseNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 5.174 ms | 2 - 193 MB | NPU
108
- | BiseNet | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 13.396 ms | 2 - 196 MB | NPU
109
- | BiseNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 4.3 ms | 2 - 195 MB | NPU
110
- | BiseNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 20.689 ms | 31 - 287 MB | NPU
111
- | BiseNet | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 105.688 ms | 32 - 246 MB | NPU
112
- | BiseNet | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 28.226 ms | 32 - 34 MB | NPU
113
- | BiseNet | TFLITE | float | Qualcomm® SA8775P | 165.438 ms | 32 - 246 MB | NPU
114
- | BiseNet | TFLITE | float | Qualcomm® QCS9075 | 56.075 ms | 0 - 66 MB | NPU
115
- | BiseNet | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 60.297 ms | 30 - 302 MB | NPU
116
- | BiseNet | TFLITE | float | Qualcomm® SA7255P | 105.688 ms | 32 - 246 MB | NPU
117
- | BiseNet | TFLITE | float | Qualcomm® SA8295P | 44.016 ms | 32 - 246 MB | NPU
118
- | BiseNet | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 15.665 ms | 31 - 255 MB | NPU
119
- | BiseNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 12.839 ms | 30 - 309 MB | NPU
120
- | BiseNet | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 8.859 ms | 7 - 240 MB | NPU
121
- | BiseNet | TFLITE | w8a8 | Qualcomm® QCS6490 | 47.175 ms | 6 - 30 MB | NPU
122
- | BiseNet | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 20.874 ms | 8 - 190 MB | NPU
123
- | BiseNet | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 12.14 ms | 8 - 10 MB | NPU
124
- | BiseNet | TFLITE | w8a8 | Qualcomm® SA8775P | 40.472 ms | 8 - 190 MB | NPU
125
- | BiseNet | TFLITE | w8a8 | Qualcomm® QCS9075 | 13.169 ms | 7 - 31 MB | NPU
126
- | BiseNet | TFLITE | w8a8 | Qualcomm® QCM6690 | 69.854 ms | 8 - 211 MB | NPU
127
- | BiseNet | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 16.192 ms | 8 - 238 MB | NPU
128
- | BiseNet | TFLITE | w8a8 | Qualcomm® SA7255P | 20.874 ms | 8 - 190 MB | NPU
129
- | BiseNet | TFLITE | w8a8 | Qualcomm® SA8295P | 15.319 ms | 8 - 193 MB | NPU
130
- | BiseNet | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 6.675 ms | 7 - 199 MB | NPU
131
- | BiseNet | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 15.975 ms | 0 - 198 MB | NPU
132
- | BiseNet | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 5.526 ms | 6 - 201 MB | NPU
133
-
134
- ## License
135
- * The license for the original implementation of BiseNet can be found
136
- [here](https://github.com/ooooverflow/BiSeNet/pull/45/files).
137
-
138
- ## References
139
- * [BiSeNet Bilateral Segmentation Network for Real-time Semantic Segmentation](https://arxiv.org/abs/1808.00897)
140
- * [Source Model Implementation](https://github.com/ooooverflow/BiSeNet)
141
-
142
- ## Community
143
- * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
144
- * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
 
2
  library_name: pytorch
3
  license: other
4
  tags:
5
+ - deprecated
6
+ pipeline_tag: other
 
7
 
8
  ---
9
+ This model is deprecated. Please refer to https://aihub.qualcomm.com for the latest models and updates.