Shufflenet-v2: Optimized for Qualcomm Devices
ShufflenetV2 is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.
This is based on the implementation of Shufflenet-v2 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 2.42, ONNX Runtime 1.24.3 | Download |
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | Download |
| TFLITE | w8a8 | Universal | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit Shufflenet-v2 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 Shufflenet-v2 on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.image_classification
Model Stats:
- Model checkpoint: Imagenet
- Input resolution: 224x224
- Number of parameters: 1.37M
- Model size (float): 5.24 MB
- Model size (w8a8): 1.47 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| Shufflenet-v2 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.426 ms | 0 - 31 MB | NPU |
| Shufflenet-v2 | ONNX | float | Snapdragon® 8 Elite Mobile | 0.437 ms | 0 - 26 MB | NPU |
| Shufflenet-v2 | ONNX | float | Snapdragon® X2 Elite | 0.424 ms | 0 - 0 MB | NPU |
| Shufflenet-v2 | ONNX | float | Snapdragon® X Elite | 0.957 ms | 2 - 2 MB | NPU |
| Shufflenet-v2 | ONNX | float | Snapdragon® X Elite | 0.957 ms | 2 - 2 MB | NPU |
| Shufflenet-v2 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.517 ms | 0 - 39 MB | NPU |
| Shufflenet-v2 | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.817 ms | 0 - 5 MB | NPU |
| Shufflenet-v2 | ONNX | float | Qualcomm® QCS9075 | 0.993 ms | 1 - 3 MB | NPU |
| Shufflenet-v2 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.437 ms | 0 - 26 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.344 ms | 0 - 31 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 0.353 ms | 0 - 32 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.342 ms | 0 - 0 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® X Elite | 0.704 ms | 0 - 0 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® X Elite | 0.704 ms | 0 - 0 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.408 ms | 0 - 36 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Qualcomm® QCS6490 | 2.702 ms | 4 - 7 MB | CPU |
| Shufflenet-v2 | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.576 ms | 0 - 4 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Qualcomm® QCS9075 | 0.688 ms | 0 - 3 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Qualcomm® QCM6690 | 2.186 ms | 0 - 9 MB | CPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.353 ms | 0 - 32 MB | NPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.449 ms | 0 - 9 MB | CPU |
| Shufflenet-v2 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.449 ms | 0 - 9 MB | CPU |
| Shufflenet-v2 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.279 ms | 1 - 30 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.362 ms | 1 - 30 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Snapdragon® X2 Elite | 0.388 ms | 1 - 1 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Snapdragon® X Elite | 0.934 ms | 1 - 1 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Snapdragon® X Elite | 0.934 ms | 1 - 1 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.501 ms | 0 - 38 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 1.721 ms | 1 - 26 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 0.782 ms | 1 - 9 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® SA8775P | 0.995 ms | 1 - 29 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® SA8775P | 0.995 ms | 1 - 29 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® SA8775P | 0.995 ms | 1 - 29 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® QCS9075 | 0.901 ms | 1 - 3 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 1.363 ms | 0 - 40 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® SA7255P | 1.721 ms | 1 - 26 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Qualcomm® SA8295P | 1.224 ms | 0 - 25 MB | NPU |
| Shufflenet-v2 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.362 ms | 1 - 30 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.209 ms | 0 - 26 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 0.265 ms | 0 - 22 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.306 ms | 0 - 0 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.588 ms | 0 - 0 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.588 ms | 0 - 0 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.338 ms | 0 - 32 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 1.118 ms | 0 - 2 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.073 ms | 0 - 24 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.469 ms | 0 - 2 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.642 ms | 0 - 25 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.642 ms | 0 - 25 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.642 ms | 0 - 25 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.581 ms | 2 - 4 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 1.338 ms | 0 - 23 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.541 ms | 0 - 33 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.073 ms | 0 - 24 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.834 ms | 0 - 21 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.265 ms | 0 - 22 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.458 ms | 0 - 22 MB | NPU |
| Shufflenet-v2 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.458 ms | 0 - 22 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.282 ms | 0 - 30 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.368 ms | 0 - 27 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.497 ms | 0 - 39 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 2.519 ms | 0 - 25 MB | CPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 0.776 ms | 0 - 3 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® SA8775P | 1.035 ms | 0 - 29 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® SA8775P | 1.035 ms | 0 - 29 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® SA8775P | 1.035 ms | 0 - 29 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® QCS9075 | 0.895 ms | 0 - 6 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 1.371 ms | 0 - 39 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® SA7255P | 2.519 ms | 0 - 25 MB | CPU |
| Shufflenet-v2 | TFLITE | float | Qualcomm® SA8295P | 1.24 ms | 0 - 24 MB | NPU |
| Shufflenet-v2 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.368 ms | 0 - 27 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.255 ms | 0 - 27 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 0.27 ms | 0 - 28 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.316 ms | 0 - 32 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® QCS6490 | 0.822 ms | 0 - 4 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.054 ms | 0 - 22 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.457 ms | 0 - 2 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® SA8775P | 0.65 ms | 0 - 26 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® SA8775P | 0.65 ms | 0 - 26 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® SA8775P | 0.65 ms | 0 - 26 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.59 ms | 0 - 3 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® QCM6690 | 1.078 ms | 0 - 23 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.517 ms | 0 - 33 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® SA7255P | 1.054 ms | 0 - 22 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Qualcomm® SA8295P | 0.807 ms | 0 - 21 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.27 ms | 0 - 28 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.443 ms | 0 - 23 MB | NPU |
| Shufflenet-v2 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.443 ms | 0 - 23 MB | NPU |
License
- The license for the original implementation of Shufflenet-v2 can be found here.
References
- ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design
- Source Model Implementation
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.
