Hugging Face's logo

qualcomm
/
RepViT

Image Classification
PyTorch
backbone
android
Model card Files
xet
 Community
qaihm-bot commited on
Commit
ed92c2c
·
verified ·
1 Parent(s): f5d281c

v0.61.0

Browse files

See https://github.com/qualcomm/ai-hub-models/releases/v0.61.0 for changelog.

Files changed (3) hide show
  1. LICENSE +1 -0
  2. README.md +118 -0
  3. release_assets.json +28 -0
LICENSE ADDED
@@ -0,0 +1 @@
 
 
1
+ The license of the original trained model can be found at https://github.com/THU-MIG/RepViT/blob/main/LICENSE.
README.md ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ license: apache-2.0
4
+ tags:
5
+ - backbone
6
+ - android
7
+ pipeline_tag: image-classification
8
+
9
+ ---
10
+
11
+ ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/repvit/web-assets/model_demo.png)
12
+
13
+ # RepViT: Optimized for Qualcomm Devices
14
+
15
+ RepViT is a lightweight pure CNN model for mobile devices, incorporating efficient architectural designs from Vision Transformers into CNNs. It achieves over 80% top-1 accuracy on ImageNet with 1ms latency on an iPhone 12.
16
+
17
+ This is based on the implementation of RepViT found [here](https://github.com/THU-MIG/RepViT).
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/v0.61.0/src/qai_hub_models/models/repvit) 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 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/repvit/releases/v0.61.0/repvit-onnx-float.zip)
32
+ | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/repvit/releases/v0.61.0/repvit-qnn_dlc-float.zip)
33
+ | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/repvit/releases/v0.61.0/repvit-tflite-float.zip)
34
+
35
+ For more device-specific assets and performance metrics, visit **[RepViT on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/repvit)**.
36
+
37
+
38
+ ### Option 2: Export with Custom Configurations
39
+
40
+ Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.61.0/src/qai_hub_models/models/repvit) Python library to compile and export the model with your own:
41
+ - Custom weights (e.g., fine-tuned checkpoints)
42
+ - Custom input shapes
43
+ - Target device and runtime configurations
44
+
45
+ This option is ideal if you need to customize the model beyond the default configuration provided here.
46
+
47
+ See our repository for [RepViT on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.61.0/src/qai_hub_models/models/repvit) for usage instructions.
48
+
49
+ ## Model Details
50
+
51
+ **Model Type:** Model_use_case.image_classification
52
+
53
+ **Model Stats:**
54
+ - Input resolution: 224x224
55
+ - Model checkpoint: Imagenet
56
+ - Model size (float): 91 MB
57
+ - Number of parameters: 22.9M
58
+
59
+ ## Performance Summary
60
+ | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
61
+ |---|---|---|---|---|---|---
62
+ | RepViT | ONNX | float | Snapdragon® X2 Elite | 1.703 ms | 2 - 2 MB | NPU
63
+ | RepViT | ONNX | float | Snapdragon® X Elite | 3.352 ms | 47 - 47 MB | NPU
64
+ | RepViT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 2.378 ms | 0 - 196 MB | NPU
65
+ | RepViT | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 11.525 ms | 1 - 193 MB | NPU
66
+ | RepViT | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 4.617 ms | 0 - 4 MB | NPU
67
+ | RepViT | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.182 ms | 0 - 232 MB | NPU
68
+ | RepViT | ONNX | float | Qualcomm® QCS8450 | 11.525 ms | 1 - 193 MB | NPU
69
+ | RepViT | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 4.872 ms | 0 - 4 MB | NPU
70
+ | RepViT | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.352 ms | 47 - 47 MB | NPU
71
+ | RepViT | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 1.814 ms | 0 - 114 MB | NPU
72
+ | RepViT | ONNX | float | Snapdragon® 8 Elite Mobile | 1.814 ms | 0 - 114 MB | NPU
73
+ | RepViT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.453 ms | 0 - 114 MB | NPU
74
+ | RepViT | QNN_DLC | float | Snapdragon® X2 Elite | 2.103 ms | 1 - 1 MB | NPU
75
+ | RepViT | QNN_DLC | float | Snapdragon® X Elite | 3.861 ms | 1 - 1 MB | NPU
76
+ | RepViT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.545 ms | 0 - 177 MB | NPU
77
+ | RepViT | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 12.232 ms | 1 - 177 MB | NPU
78
+ | RepViT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 4.764 ms | 1 - 4 MB | NPU
79
+ | RepViT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 14.457 ms | 1 - 101 MB | NPU
80
+ | RepViT | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.467 ms | 1 - 19 MB | NPU
81
+ | RepViT | QNN_DLC | float | Qualcomm® SA8775P | 4.83 ms | 1 - 103 MB | NPU
82
+ | RepViT | QNN_DLC | float | Qualcomm® SA8650P | 4.83 ms | 1 - 103 MB | NPU
83
+ | RepViT | QNN_DLC | float | Qualcomm® SA8255P | 4.83 ms | 1 - 103 MB | NPU
84
+ | RepViT | QNN_DLC | float | Qualcomm® QCS8450 | 12.232 ms | 1 - 177 MB | NPU
85
+ | RepViT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 4.578 ms | 3 - 5 MB | NPU
86
+ | RepViT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.861 ms | 1 - 1 MB | NPU
87
+ | RepViT | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 1.909 ms | 0 - 104 MB | NPU
88
+ | RepViT | QNN_DLC | float | Qualcomm® SA7255P | 14.457 ms | 1 - 101 MB | NPU
89
+ | RepViT | QNN_DLC | float | Qualcomm® SA8295P | 9.678 ms | 1 - 106 MB | NPU
90
+ | RepViT | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 1.909 ms | 0 - 104 MB | NPU
91
+ | RepViT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.508 ms | 1 - 104 MB | NPU
92
+ | RepViT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.546 ms | 0 - 234 MB | NPU
93
+ | RepViT | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 12.232 ms | 0 - 228 MB | NPU
94
+ | RepViT | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 4.776 ms | 0 - 51 MB | NPU
95
+ | RepViT | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 14.489 ms | 0 - 162 MB | NPU
96
+ | RepViT | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.451 ms | 0 - 2 MB | NPU
97
+ | RepViT | TFLITE | float | Qualcomm® SA8775P | 4.868 ms | 0 - 164 MB | NPU
98
+ | RepViT | TFLITE | float | Qualcomm® SA8650P | 4.868 ms | 0 - 164 MB | NPU
99
+ | RepViT | TFLITE | float | Qualcomm® SA8255P | 4.868 ms | 0 - 164 MB | NPU
100
+ | RepViT | TFLITE | float | Qualcomm® QCS8450 | 12.232 ms | 0 - 228 MB | NPU
101
+ | RepViT | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 4.588 ms | 0 - 50 MB | NPU
102
+ | RepViT | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 1.89 ms | 0 - 164 MB | NPU
103
+ | RepViT | TFLITE | float | Qualcomm® SA7255P | 14.489 ms | 0 - 162 MB | NPU
104
+ | RepViT | TFLITE | float | Qualcomm® SA8295P | 9.653 ms | 0 - 161 MB | NPU
105
+ | RepViT | TFLITE | float | Snapdragon® 8 Elite Mobile | 1.89 ms | 0 - 164 MB | NPU
106
+ | RepViT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.508 ms | 0 - 167 MB | NPU
107
+
108
+ ## License
109
+ * The license for the original implementation of RepViT can be found
110
+ [here](https://github.com/THU-MIG/RepViT/blob/main/LICENSE).
111
+
112
+ ## References
113
+ * [RepViT: Revisiting Mobile CNN From ViT Perspective](https://arxiv.org/abs/2307.09283)
114
+ * [Source Model Implementation](https://github.com/THU-MIG/RepViT)
115
+
116
+ ## Community
117
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
118
+ * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
release_assets.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "version": "0.61.0",
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/repvit/releases/v0.61.0/repvit-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/repvit/releases/v0.61.0/repvit-qnn_dlc-float.zip"
18
+ },
19
+ "tflite": {
20
+ "tool_versions": {
21
+ "qairt": "2.45.0.260326154327"
22
+ },
23
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/repvit/releases/v0.61.0/repvit-tflite-float.zip"
24
+ }
25
+ }
26
+ }
27
+ }
28
+ }