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. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.
Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up 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 2.50, ONNX Runtime 1.30.0 | Download |
| ONNX | w8a8 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.50 | Download |
| QNN_DLC | w8a8 | Universal | QAIRT 2.50 | Download |
| TFLITE | float | Universal | QAIRT 2.50 | Download |
| TFLITE | w8a8 | Universal | QAIRT 2.50 | Download |
For more device-specific assets and performance metrics, visit ResNet-3D on Qualcomm® AI Hub.
Option 2: Export with Custom Configurations
Use the Qualcomm® AI Hub Models 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 for usage instructions.
Model Details
Model Type: Model_use_case.video_classification
Model Stats:
- Input resolution: 112x112
- Model checkpoint: Kinetics-400
- Model size (float): 127 MB
- Model size (w8a8): 32.1 MB
- Number of parameters: 33.4M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| ResNet-3D | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 5.976 ms | 2 - 179 MB | NPU |
| ResNet-3D | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.663 ms | 0 - 172 MB | NPU |
| ResNet-3D | ONNX | float | Snapdragon® X2 Elite | 6.913 ms | 2 - 2 MB | NPU |
| ResNet-3D | ONNX | float | Snapdragon® X Elite | 13.125 ms | 64 - 64 MB | NPU |
| ResNet-3D | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 9.387 ms | 2 - 231 MB | NPU |
| ResNet-3D | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 23.341 ms | 2 - 216 MB | NPU |
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 26.869 ms | 2 - 9 MB | NPU |
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 12.91 ms | 2 - 5 MB | NPU |
| ResNet-3D | ONNX | float | Qualcomm® QCS8450 | 23.341 ms | 2 - 216 MB | NPU |
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 24.541 ms | 2 - 8 MB | NPU |
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 13.125 ms | 64 - 64 MB | NPU |
| ResNet-3D | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 7.663 ms | 0 - 172 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2.03 ms | 0 - 164 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 2.664 ms | 0 - 167 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Snapdragon® X2 Elite | 2.216 ms | 2 - 2 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Snapdragon® X Elite | 4.429 ms | 33 - 33 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 3.207 ms | 0 - 231 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 6.15 ms | 1 - 229 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 18.91 ms | 0 - 4 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 4.85 ms | 0 - 5 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 4.336 ms | 0 - 41 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® QCS8450 | 6.15 ms | 1 - 229 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 4.241 ms | 0 - 4 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 4.429 ms | 33 - 33 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 91.409 ms | 0 - 172 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 7.808 ms | 0 - 161 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 2.664 ms | 0 - 167 MB | NPU |
| ResNet-3D | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 7.808 ms | 0 - 161 MB | NPU |
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 5.87 ms | 2 - 180 MB | NPU |
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.678 ms | 2 - 178 MB | NPU |
| ResNet-3D | QNN_DLC | float | Snapdragon® X2 Elite | 7.525 ms | 2 - 2 MB | NPU |
| ResNet-3D | QNN_DLC | float | Snapdragon® X Elite | 13.328 ms | 2 - 2 MB | NPU |
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 9.351 ms | 0 - 227 MB | NPU |
| ResNet-3D | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 23.428 ms | 2 - 218 MB | NPU |
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 26.755 ms | 2 - 7 MB | NPU |
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 12.85 ms | 2 - 4 MB | NPU |
| ResNet-3D | QNN_DLC | float | Qualcomm® QCS8450 | 23.428 ms | 2 - 218 MB | NPU |
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 23.724 ms | 2 - 7 MB | NPU |
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 13.328 ms | 2 - 2 MB | NPU |
| ResNet-3D | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 7.678 ms | 2 - 178 MB | NPU |
| ResNet-3D | QNN_DLC | float | Qualcomm® SA8295P | 26.053 ms | 0 - 164 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.983 ms | 1 - 160 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 2.699 ms | 1 - 158 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 2.527 ms | 1 - 1 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® X Elite | 4.661 ms | 1 - 1 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 3.276 ms | 0 - 230 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 5.736 ms | 1 - 229 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 19.07 ms | 1 - 3 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 4.87 ms | 1 - 4 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 4.328 ms | 1 - 2 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 5.736 ms | 1 - 229 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 4.203 ms | 0 - 3 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 4.661 ms | 1 - 1 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 88.505 ms | 1 - 174 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 7.879 ms | 1 - 159 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 2.699 ms | 1 - 158 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® SA7255P | 15.057 ms | 1 - 167 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Qualcomm® SA8295P | 7.994 ms | 0 - 173 MB | NPU |
| ResNet-3D | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 7.879 ms | 1 - 159 MB | NPU |
| ResNet-3D | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 158.32 ms | 0 - 204 MB | NPU |
| ResNet-3D | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 161.693 ms | 0 - 201 MB | NPU |
| ResNet-3D | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 173.211 ms | 0 - 252 MB | NPU |
| ResNet-3D | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 262.989 ms | 1 - 249 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 255.002 ms | 0 - 71 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 230.165 ms | 0 - 3 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® SA8775P | 230.843 ms | 0 - 199 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® SA8650P | 230.843 ms | 0 - 199 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® SA8255P | 230.843 ms | 0 - 199 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® QCS8450 | 262.989 ms | 1 - 249 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 241.562 ms | 0 - 70 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 161.693 ms | 0 - 201 MB | NPU |
| ResNet-3D | TFLITE | float | Qualcomm® SA8295P | 288.544 ms | 0 - 194 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 150.96 ms | 0 - 279 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 127.045 ms | 0 - 242 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 181.167 ms | 0 - 297 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 261.752 ms | 1 - 297 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 1724.309 ms | 695 - 1060 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 240.762 ms | 0 - 69 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 238.4 ms | 0 - 3 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA8650P | 887.812 ms | 692 - 707 MB | CPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA8255P | 887.812 ms | 692 - 707 MB | CPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® QCS8450 | 261.752 ms | 1 - 297 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 241.708 ms | 0 - 69 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 1480.445 ms | 650 - 799 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1244.86 ms | 691 - 751 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 127.045 ms | 0 - 242 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Qualcomm® SA8295P | 258.048 ms | 0 - 237 MB | NPU |
| ResNet-3D | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1244.86 ms | 691 - 751 MB | NPU |
License
- The license for the original implementation of ResNet-3D can be found here.
References
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.
