|
|
--- |
|
|
library_name: pytorch |
|
|
license: other |
|
|
tags: |
|
|
- android |
|
|
pipeline_tag: image-to-image |
|
|
|
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# QuickSRNetLarge: Optimized for Qualcomm Devices |
|
|
|
|
|
QuickSRNet Large is designed for upscaling images on mobile platforms to sharpen in real-time. |
|
|
|
|
|
This is based on the implementation of QuickSRNetLarge found [here](https://github.com/quic/aimet-model-zoo/tree/develop/aimet_zoo_torch/quicksrnet). |
|
|
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/quicksrnetlarge) 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/quicksrnetlarge/releases/v0.46.0/quicksrnetlarge-onnx-float.zip) |
|
|
| QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/quicksrnetlarge/releases/v0.46.0/quicksrnetlarge-qnn_dlc-float.zip) |
|
|
| QNN_DLC | w8a8 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/quicksrnetlarge/releases/v0.46.0/quicksrnetlarge-qnn_dlc-w8a8.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/quicksrnetlarge/releases/v0.46.0/quicksrnetlarge-tflite-float.zip) |
|
|
| TFLITE | w8a8 | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/quicksrnetlarge/releases/v0.46.0/quicksrnetlarge-tflite-w8a8.zip) |
|
|
|
|
|
For more device-specific assets and performance metrics, visit **[QuickSRNetLarge on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/quicksrnetlarge)**. |
|
|
|
|
|
|
|
|
### 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/quicksrnetlarge) 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 [QuickSRNetLarge on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/quicksrnetlarge) for usage instructions. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
**Model Type:** Model_use_case.super_resolution |
|
|
|
|
|
**Model Stats:** |
|
|
- Model checkpoint: quicksrnet_large_3x_checkpoint |
|
|
- Input resolution: 128x128 |
|
|
- Number of parameters: 436K |
|
|
- Model size (float): 1.67 MB |
|
|
- Model size (w8a8): 462 KB |
|
|
|
|
|
## Performance Summary |
|
|
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|
|
|---|---|---|---|---|---|--- |
|
|
| QuickSRNetLarge | ONNX | float | Snapdragon® X Elite | 2.289 ms | 7 - 7 MB | NPU |
|
|
| QuickSRNetLarge | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.721 ms | 0 - 103 MB | NPU |
|
|
| QuickSRNetLarge | ONNX | float | Qualcomm® QCS8550 (Proxy) | 2.355 ms | 0 - 59 MB | NPU |
|
|
| QuickSRNetLarge | ONNX | float | Qualcomm® QCS9075 | 3.723 ms | 7 - 10 MB | NPU |
|
|
| QuickSRNetLarge | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.354 ms | 0 - 92 MB | NPU |
|
|
| QuickSRNetLarge | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.054 ms | 0 - 91 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Snapdragon® X Elite | 2.019 ms | 0 - 0 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.29 ms | 0 - 37 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 11.865 ms | 0 - 22 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 1.825 ms | 0 - 1 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Qualcomm® SA8775P | 14.61 ms | 0 - 25 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Qualcomm® QCS9075 | 3.349 ms | 0 - 5 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 3.288 ms | 0 - 38 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Qualcomm® SA7255P | 11.865 ms | 0 - 22 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Qualcomm® SA8295P | 3.937 ms | 0 - 20 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.04 ms | 0 - 22 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.831 ms | 0 - 27 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.738 ms | 0 - 0 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.395 ms | 0 - 32 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 2.713 ms | 2 - 4 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.947 ms | 0 - 22 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.589 ms | 0 - 1 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.771 ms | 0 - 23 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.832 ms | 2 - 4 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 8.347 ms | 0 - 134 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.857 ms | 0 - 32 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.947 ms | 0 - 22 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Qualcomm® SA8295P | 1.343 ms | 0 - 20 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.345 ms | 0 - 21 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.884 ms | 0 - 21 MB | NPU |
|
|
| QuickSRNetLarge | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.266 ms | 0 - 24 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.461 ms | 0 - 40 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 12.085 ms | 3 - 29 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 2.154 ms | 0 - 1 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Qualcomm® SA8775P | 3.734 ms | 0 - 26 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Qualcomm® QCS9075 | 3.697 ms | 3 - 9 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 3.514 ms | 0 - 40 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Qualcomm® SA7255P | 12.085 ms | 3 - 29 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Qualcomm® SA8295P | 4.177 ms | 0 - 22 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.094 ms | 0 - 24 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.872 ms | 0 - 28 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.483 ms | 0 - 34 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® QCS6490 | 2.695 ms | 0 - 3 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 2.228 ms | 0 - 26 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.766 ms | 0 - 1 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® SA8775P | 0.994 ms | 0 - 27 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.965 ms | 0 - 3 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® QCM6690 | 8.51 ms | 0 - 136 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 1.034 ms | 0 - 35 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® SA7255P | 2.228 ms | 0 - 26 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Qualcomm® SA8295P | 1.476 ms | 0 - 23 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.371 ms | 0 - 26 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.977 ms | 0 - 23 MB | NPU |
|
|
| QuickSRNetLarge | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.331 ms | 0 - 26 MB | NPU |
|
|
|
|
|
## License |
|
|
* The license for the original implementation of QuickSRNetLarge can be found |
|
|
[here](https://github.com/quic/aimet-model-zoo/blob/develop/LICENSE.pdf). |
|
|
|
|
|
## References |
|
|
* [QuickSRNet: Plain Single-Image Super-Resolution Architecture for Faster Inference on Mobile Platforms](https://arxiv.org/abs/2303.04336) |
|
|
* [Source Model Implementation](https://github.com/quic/aimet-model-zoo/tree/develop/aimet_zoo_torch/quicksrnet) |
|
|
|
|
|
## 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). |
|
|
|
