Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data
Paper
• 2107.10833 • Published
• 1
Real-ESRGAN-x4plus: Optimized for Qualcomm Devices
Real-ESRGAN is a machine learning model that upscales an image with minimal loss in quality. The implementation is a derivative of the Real-ESRGAN-x4plus architecture, a larger and more powerful version compared to the Real-ESRGAN-general-x4v3 architecture.
This is based on the implementation of Real-ESRGAN-x4plus 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 1.24.1 | Download |
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| QNN_DLC | float | Universal | QAIRT 2.43 | Download |
| QNN_DLC | w8a8 | Universal | QAIRT 2.43 | Download |
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | Download |
| TFLITE | w8a8 | Universal | QAIRT 2.43, TFLite 2.17.0 | Download |
For more device-specific assets and performance metrics, visit Real-ESRGAN-x4plus 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 Real-ESRGAN-x4plus on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.super_resolution
Model Stats:
- Model checkpoint: RealESRGAN_x4plus
- Input resolution: 128x128
- Number of parameters: 16.7M
- Model size (float): 63.9 MB
- Model size (w8a8): 16.7 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® X Elite | 26.399 ms | 21 - 21 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 17.141 ms | 2 - 787 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 24.98 ms | 0 - 30 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Qualcomm® QCS9075 | 31.299 ms | 2 - 4 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 12.69 ms | 2 - 486 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 9.146 ms | 2 - 497 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® X2 Elite | 10.302 ms | 23 - 23 MB | NPU |
License
- The license for the original implementation of Real-ESRGAN-x4plus can be found here.
References
- Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data
- 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.