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 | float | Snapdragon® 8 Elite Gen 5 Mobile | 27.434 ms | 7 - 353 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | float | Snapdragon® X2 Elite | 34.513 ms | 37 - 37 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | float | Snapdragon® X Elite | 65.552 ms | 37 - 37 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 51.166 ms | 0 - 785 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | float | Qualcomm® QCS8550 (Proxy) | 63.612 ms | 0 - 48 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | float | Qualcomm® QCS9075 | 106.809 ms | 8 - 11 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 38.309 ms | 7 - 352 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 9.145 ms | 2 - 496 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® X2 Elite | 10.322 ms | 23 - 23 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® X Elite | 26.241 ms | 21 - 21 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 17.037 ms | 2 - 783 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 24.967 ms | 0 - 30 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Qualcomm® QCS9075 | 30.866 ms | 2 - 4 MB | NPU |
| Real-ESRGAN-x4plus | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 12.658 ms | 2 - 484 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 25.784 ms | 0 - 327 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Snapdragon® X2 Elite | 34.407 ms | 0 - 0 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Snapdragon® X Elite | 64.972 ms | 0 - 0 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 48.691 ms | 0 - 750 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 452.192 ms | 0 - 348 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 63.027 ms | 0 - 3 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Qualcomm® SA8775P | 105.543 ms | 0 - 347 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Qualcomm® QCS9075 | 108.557 ms | 0 - 5 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 116.955 ms | 0 - 755 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Qualcomm® SA7255P | 452.192 ms | 0 - 348 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Qualcomm® SA8295P | 111.244 ms | 0 - 357 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 38.141 ms | 0 - 329 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 9.404 ms | 0 - 430 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 10.833 ms | 0 - 0 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Snapdragon® X Elite | 25.739 ms | 0 - 0 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 16.708 ms | 0 - 699 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 109.664 ms | 0 - 2 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 68.857 ms | 0 - 476 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 24.378 ms | 0 - 434 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® SA8775P | 22.928 ms | 0 - 476 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 30.155 ms | 0 - 2 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 481.093 ms | 0 - 466 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 37.955 ms | 0 - 728 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® SA7255P | 68.857 ms | 0 - 476 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Qualcomm® SA8295P | 36.24 ms | 0 - 485 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 12.369 ms | 0 - 417 MB | NPU |
| Real-ESRGAN-x4plus | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 38.741 ms | 0 - 483 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 27.968 ms | 3 - 371 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 50.86 ms | 0 - 795 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 452.345 ms | 3 - 392 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 61.726 ms | 3 - 8 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Qualcomm® SA8775P | 505.009 ms | 3 - 391 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 144.717 ms | 4 - 783 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Qualcomm® SA7255P | 452.345 ms | 3 - 392 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Qualcomm® SA8295P | 110.479 ms | 3 - 395 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 38.123 ms | 3 - 365 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 7.831 ms | 1 - 502 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 13.801 ms | 1 - 672 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® QCS6490 | 93.007 ms | 1 - 26 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 64.873 ms | 1 - 446 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 18.695 ms | 1 - 4 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® SA8775P | 18.452 ms | 1 - 448 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® QCS9075 | 25.433 ms | 1 - 27 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® QCM6690 | 440.555 ms | 1 - 441 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 32.344 ms | 1 - 680 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® SA7255P | 64.873 ms | 1 - 446 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Qualcomm® SA8295P | 34.946 ms | 1 - 444 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 11.678 ms | 1 - 476 MB | NPU |
| Real-ESRGAN-x4plus | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 32.581 ms | 1 - 451 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.