|
|
--- |
|
|
library_name: pytorch |
|
|
license: other |
|
|
tags: |
|
|
- backbone |
|
|
- bu_auto |
|
|
- real_time |
|
|
- android |
|
|
pipeline_tag: image-classification |
|
|
|
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# MobileNet-v3-Large: Optimized for Qualcomm Devices |
|
|
|
|
|
MobileNet-v3-Large 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 MobileNet-v3-Large found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/mobilenetv3.py). |
|
|
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v3_large) 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 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_large/releases/v0.46.0/mobilenet_v3_large-onnx-float.zip) |
|
|
| ONNX | w8a16 | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_large/releases/v0.46.0/mobilenet_v3_large-onnx-w8a16.zip) |
|
|
| QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_large/releases/v0.46.0/mobilenet_v3_large-qnn_dlc-float.zip) |
|
|
| QNN_DLC | w8a16 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_large/releases/v0.46.0/mobilenet_v3_large-qnn_dlc-w8a16.zip) |
|
|
| TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v3_large/releases/v0.46.0/mobilenet_v3_large-tflite-float.zip) |
|
|
|
|
|
For more device-specific assets and performance metrics, visit **[MobileNet-v3-Large on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/mobilenet_v3_large)**. |
|
|
|
|
|
|
|
|
### Option 2: Export with Custom Configurations |
|
|
|
|
|
Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v3_large) 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 [MobileNet-v3-Large on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v3_large) for usage instructions. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
**Model Type:** Model_use_case.image_classification |
|
|
|
|
|
**Model Stats:** |
|
|
- Model checkpoint: Imagenet |
|
|
- Input resolution: 224x224 |
|
|
- Number of parameters: 5.47M |
|
|
- Model size (float): 20.9 MB |
|
|
- Model size (w8a16): 6.35 MB |
|
|
|
|
|
## Performance Summary |
|
|
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|
|
|---|---|---|---|---|---|--- |
|
|
| MobileNet-v3-Large | ONNX | float | Snapdragon® X Elite | 0.826 ms | 13 - 13 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.682 ms | 0 - 123 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.901 ms | 0 - 25 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | float | Qualcomm® QCS9075 | 1.257 ms | 1 - 3 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.528 ms | 0 - 103 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.479 ms | 0 - 104 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Snapdragon® X Elite | 0.824 ms | 6 - 6 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.674 ms | 0 - 127 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Qualcomm® QCS6490 | 55.392 ms | 21 - 26 MB | CPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.918 ms | 0 - 9 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Qualcomm® QCS9075 | 1.003 ms | 0 - 3 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Qualcomm® QCM6690 | 30.839 ms | 24 - 33 MB | CPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.483 ms | 0 - 109 MB | NPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 21.181 ms | 24 - 32 MB | CPU |
|
|
| MobileNet-v3-Large | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.397 ms | 0 - 107 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Snapdragon® X Elite | 1.183 ms | 1 - 1 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.683 ms | 0 - 54 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 2.975 ms | 1 - 35 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 1.003 ms | 1 - 2 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Qualcomm® SA8775P | 1.346 ms | 1 - 37 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Qualcomm® QCS9075 | 1.227 ms | 1 - 3 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 1.924 ms | 0 - 58 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Qualcomm® SA7255P | 2.975 ms | 1 - 35 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Qualcomm® SA8295P | 1.823 ms | 0 - 35 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.523 ms | 0 - 39 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.424 ms | 1 - 40 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Snapdragon® X Elite | 1.115 ms | 0 - 0 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.659 ms | 0 - 47 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 2.7 ms | 2 - 4 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 2.077 ms | 0 - 34 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.941 ms | 0 - 7 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® SA8775P | 1.15 ms | 0 - 35 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 1.163 ms | 2 - 4 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 4.074 ms | 0 - 146 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 1.244 ms | 0 - 52 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® SA7255P | 2.077 ms | 0 - 34 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Qualcomm® SA8295P | 1.542 ms | 0 - 31 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.456 ms | 0 - 31 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 0.992 ms | 0 - 31 MB | NPU |
|
|
| MobileNet-v3-Large | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.363 ms | 0 - 35 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.682 ms | 0 - 61 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 3.053 ms | 0 - 40 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 1.011 ms | 0 - 2 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Qualcomm® SA8775P | 1.384 ms | 0 - 43 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Qualcomm® QCS9075 | 1.275 ms | 0 - 15 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 1.922 ms | 0 - 64 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Qualcomm® SA7255P | 3.053 ms | 0 - 40 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Qualcomm® SA8295P | 1.842 ms | 0 - 40 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.52 ms | 0 - 40 MB | NPU |
|
|
| MobileNet-v3-Large | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.43 ms | 0 - 45 MB | NPU |
|
|
|
|
|
## License |
|
|
* The license for the original implementation of MobileNet-v3-Large can be found |
|
|
[here](https://github.com/pytorch/vision/blob/main/LICENSE). |
|
|
|
|
|
## References |
|
|
* [Searching for MobileNetV3](https://arxiv.org/abs/1905.02244) |
|
|
* [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/mobilenetv3.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). |
|
|
|
