---
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).