Hugging Face's logo

qualcomm
/
ResNet-3D

Video Classification
PyTorch
backbone
android
Model card Files
xet
 Community
ResNet-3D
File size: 12,557 Bytes
9fe0cf4
 
87db2ee
9fe0cf4
 
 
1ac4e76
9fe0cf4
 
 
 
 
e2fc2bd
9fe0cf4
 
 
e2fc2bd
286ff78
e2fc2bd
 
 
 
 
 
 
 
 
 
 
 
286ff78
 
 
 
 
 
e2fc2bd
 
 
 
 
 
286ff78
e2fc2bd
 
 
 
 
 
286ff78
e2fc2bd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
286ff78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9fe0cf4
 
80acba8
 
9fe0cf4
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
---
library_name: pytorch
license: other
tags:
- backbone
- android
pipeline_tag: video-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet_3d/web-assets/model_demo.png)

# ResNet-3D: Optimized for Qualcomm Devices

ResNet 3D is a network with 3D convolutions used for video understanding.

This is based on the implementation of ResNet-3D found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/video/resnet.py).
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/resnet_3d) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

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.

## Getting Started
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| 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/resnet_3d/releases/v0.59.0/resnet_3d-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet_3d/releases/v0.59.0/resnet_3d-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet_3d/releases/v0.59.0/resnet_3d-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet_3d/releases/v0.59.0/resnet_3d-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet_3d/releases/v0.59.0/resnet_3d-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet_3d/releases/v0.59.0/resnet_3d-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[ResNet-3D on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/resnet_3d)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/resnet_3d) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [ResNet-3D on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/resnet_3d) for usage instructions.

## Model Details

**Model Type:** Model_use_case.video_classification

**Model Stats:**
- Model checkpoint: Kinetics-400
- Input resolution: 112x112
- Number of parameters: 33.4M
- Model size (float): 127 MB
- Model size (w8a8): 32.1 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| ResNet-3D | ONNX | float | Snapdragon® X2 Elite | 35.768 ms | 12 - 12 MB | NPU
| ResNet-3D | ONNX | float | Snapdragon® X Elite | 67.732 ms | 58 - 58 MB | NPU
| ResNet-3D | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 50.574 ms | 3 - 586 MB | NPU
| ResNet-3D | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 135.243 ms | 12 - 502 MB | NPU
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 67.478 ms | 1 - 72 MB | NPU
| ResNet-3D | ONNX | float | Qualcomm® QCS8450 | 135.243 ms | 12 - 502 MB | NPU
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 125.727 ms | 11 - 26 MB | NPU
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 67.732 ms | 58 - 58 MB | NPU
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 41.821 ms | 1 - 391 MB | NPU
| ResNet-3D | ONNX | float | Snapdragon® 8 Elite Mobile | 41.821 ms | 1 - 391 MB | NPU
| ResNet-3D | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 30.064 ms | 1 - 408 MB | NPU
| ResNet-3D | ONNX | w8a8 | Snapdragon® X2 Elite | 11.575 ms | 3 - 3 MB | NPU
| ResNet-3D | ONNX | w8a8 | Snapdragon® X Elite | 22.394 ms | 32 - 32 MB | NPU
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 16.789 ms | 0 - 415 MB | NPU
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 35.729 ms | 0 - 416 MB | NPU
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 22.205 ms | 0 - 40 MB | NPU
| ResNet-3D | ONNX | w8a8 | Qualcomm® QCS8450 | 35.729 ms | 0 - 416 MB | NPU
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 21.375 ms | 0 - 6 MB | NPU
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 22.394 ms | 32 - 32 MB | NPU
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 14.941 ms | 0 - 292 MB | NPU
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 14.941 ms | 0 - 292 MB | NPU
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 10.163 ms | 2 - 301 MB | NPU
| ResNet-3D | QNN_DLC | float | Snapdragon® X2 Elite | 36.206 ms | 12 - 12 MB | NPU
| ResNet-3D | QNN_DLC | float | Snapdragon® X Elite | 70.807 ms | 12 - 12 MB | NPU
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 51.436 ms | 12 - 599 MB | NPU
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 134.47 ms | 7 - 499 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 467.671 ms | 0 - 395 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 69.21 ms | 12 - 15 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® SA8775P | 123.909 ms | 0 - 395 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® SA8650P | 123.909 ms | 0 - 395 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® SA8255P | 123.909 ms | 0 - 395 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® QCS8450 | 134.47 ms | 7 - 499 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 124.41 ms | 12 - 25 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 70.807 ms | 12 - 12 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 41.887 ms | 0 - 392 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® SA7255P | 467.671 ms | 0 - 395 MB | NPU
| ResNet-3D | QNN_DLC | float | Qualcomm® SA8295P | 132.99 ms | 0 - 317 MB | NPU
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 41.887 ms | 0 - 392 MB | NPU
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 29.974 ms | 12 - 407 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 17.371 ms | 3 - 3 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® X Elite | 23.518 ms | 3 - 3 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 17.096 ms | 3 - 409 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 36.48 ms | 3 - 412 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 137.143 ms | 3 - 7 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8275 | 75.932 ms | 3 - 283 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 22.876 ms | 3 - 319 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® SA8775P | 21.831 ms | 3 - 285 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® SA8650P | 21.831 ms | 3 - 285 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® SA8255P | 21.831 ms | 3 - 285 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 36.48 ms | 3 - 412 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 23.402 ms | 2 - 7 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 23.518 ms | 3 - 3 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 463.531 ms | 3 - 360 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 42.75 ms | 3 - 299 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 15.386 ms | 3 - 291 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® SA7255P | 75.932 ms | 3 - 283 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® SA8295P | 39.834 ms | 0 - 281 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 15.386 ms | 3 - 291 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 10.718 ms | 3 - 299 MB | NPU
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 42.75 ms | 3 - 299 MB | NPU
| ResNet-3D | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1128.526 ms | 0 - 642 MB | NPU
| ResNet-3D | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 1624.894 ms | 0 - 604 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 2775.179 ms | 1 - 456 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1422.381 ms | 0 - 3 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® SA8775P | 1482.157 ms | 1 - 476 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® SA8650P | 1482.157 ms | 1 - 476 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® SA8255P | 1482.157 ms | 1 - 476 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® QCS8450 | 1624.894 ms | 0 - 604 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 1417.474 ms | 0 - 83 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 955.799 ms | 0 - 470 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® SA7255P | 2775.179 ms | 1 - 456 MB | NPU
| ResNet-3D | TFLITE | float | Qualcomm® SA8295P | 1730.667 ms | 0 - 454 MB | NPU
| ResNet-3D | TFLITE | float | Snapdragon® 8 Elite Mobile | 955.799 ms | 0 - 470 MB | NPU
| ResNet-3D | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 887.67 ms | 0 - 473 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 943.378 ms | 0 - 999 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1414.361 ms | 0 - 868 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 9128.12 ms | 3462 - 5277 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8275 | 2553.225 ms | 1 - 820 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1264.192 ms | 0 - 3 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA8775P | 1260.25 ms | 1 - 819 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA8650P | 1260.25 ms | 1 - 819 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA8255P | 1260.25 ms | 1 - 819 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® QCS8450 | 1414.361 ms | 0 - 868 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1271.197 ms | 0 - 77 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 388866.684 ms | 913 - 1123 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 345054.249 ms | 920 - 1202 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 689.293 ms | 0 - 728 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA7255P | 2553.225 ms | 1 - 820 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA8295P | 1341.683 ms | 1 - 685 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 689.293 ms | 0 - 728 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 814.412 ms | 1 - 929 MB | NPU
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 345054.249 ms | 920 - 1202 MB | NPU

## License
* The license for the original implementation of ResNet-3D can be found
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).

## References
* [A Closer Look at Spatiotemporal Convolutions for Action Recognition](https://arxiv.org/abs/1711.11248)
* [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/video/resnet.py)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).