XLSR: Optimized for Qualcomm Devices
XLSR is designed for lightweight real-time upscaling of images.
This is based on the implementation of XLSR 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 2.45, ONNX Runtime 1.27.1 | Download |
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | 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 XLSR 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 XLSR on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.super_resolution
Model Stats:
- Input resolution: 128x128
- Model checkpoint: xlsr_3x_checkpoint
- Model size (float): 115 KB
- Model size (w8a8): 45.6 KB
- Number of parameters: 28.0K
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| XLSR | ONNX | float | Snapdragon® X2 Elite | 0.375 ms | 8 - 8 MB | NPU |
| XLSR | ONNX | float | Snapdragon® X Elite | 0.868 ms | 8 - 8 MB | NPU |
| XLSR | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.575 ms | 0 - 28 MB | NPU |
| XLSR | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 1.248 ms | 1 - 35 MB | NPU |
| XLSR | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 1.179 ms | 0 - 4 MB | NPU |
| XLSR | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.869 ms | 0 - 2 MB | NPU |
| XLSR | ONNX | float | Qualcomm® QCS8450 | 1.248 ms | 1 - 35 MB | NPU |
| XLSR | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 1.18 ms | 6 - 9 MB | NPU |
| XLSR | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 0.868 ms | 8 - 8 MB | NPU |
| XLSR | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.467 ms | 0 - 23 MB | NPU |
| XLSR | ONNX | float | Snapdragon® 8 Elite Mobile | 0.467 ms | 0 - 23 MB | NPU |
| XLSR | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.4 ms | 0 - 19 MB | NPU |
| XLSR | ONNX | w8a8 | Snapdragon® X2 Elite | 0.257 ms | 3 - 3 MB | NPU |
| XLSR | ONNX | w8a8 | Snapdragon® X Elite | 0.643 ms | 3 - 3 MB | NPU |
| XLSR | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.423 ms | 0 - 33 MB | NPU |
| XLSR | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.72 ms | 2 - 39 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 3.32 ms | 2 - 4 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.705 ms | 0 - 4 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.643 ms | 0 - 2 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® QCS8450 | 0.72 ms | 2 - 39 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.786 ms | 0 - 3 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 0.643 ms | 3 - 3 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 3.055 ms | 0 - 137 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.659 ms | 0 - 23 MB | NPU |
| XLSR | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.311 ms | 0 - 23 MB | NPU |
| XLSR | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 0.311 ms | 0 - 23 MB | NPU |
| XLSR | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.268 ms | 0 - 24 MB | NPU |
| XLSR | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.659 ms | 0 - 23 MB | NPU |
| XLSR | QNN_DLC | float | Snapdragon® X2 Elite | 0.727 ms | 0 - 0 MB | NPU |
| XLSR | QNN_DLC | float | Snapdragon® X Elite | 0.847 ms | 0 - 0 MB | NPU |
| XLSR | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.458 ms | 0 - 30 MB | NPU |
| XLSR | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 1.051 ms | 0 - 36 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 0.939 ms | 0 - 6 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.785 ms | 0 - 3 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® SA8775P | 1.139 ms | 0 - 22 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® SA8650P | 1.139 ms | 0 - 22 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® SA8255P | 1.139 ms | 0 - 22 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® QCS8450 | 1.051 ms | 0 - 36 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 1.033 ms | 0 - 5 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 0.847 ms | 0 - 0 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.369 ms | 0 - 26 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® SA7255P | 2.178 ms | 0 - 21 MB | NPU |
| XLSR | QNN_DLC | float | Qualcomm® SA8295P | 1.392 ms | 0 - 17 MB | NPU |
| XLSR | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.369 ms | 0 - 26 MB | NPU |
| XLSR | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.335 ms | 0 - 22 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.27 ms | 0 - 0 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.482 ms | 0 - 0 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.248 ms | 0 - 29 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.472 ms | 0 - 29 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 1.412 ms | 2 - 4 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.419 ms | 0 - 3 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.381 ms | 0 - 12 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.576 ms | 0 - 22 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8650P | 0.576 ms | 0 - 22 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8255P | 0.576 ms | 0 - 22 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 0.472 ms | 0 - 29 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.512 ms | 2 - 4 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 0.482 ms | 0 - 0 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 1.6 ms | 0 - 19 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.42 ms | 0 - 20 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.214 ms | 0 - 21 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.771 ms | 0 - 18 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 0.214 ms | 0 - 21 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.185 ms | 0 - 23 MB | NPU |
| XLSR | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.42 ms | 0 - 20 MB | NPU |
| XLSR | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.398 ms | 0 - 32 MB | NPU |
| XLSR | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 2.497 ms | 0 - 37 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 2.179 ms | 3 - 9 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.215 ms | 0 - 1 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® SA8775P | 2.655 ms | 0 - 23 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® SA8650P | 2.655 ms | 0 - 23 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® SA8255P | 2.655 ms | 0 - 23 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® QCS8450 | 2.497 ms | 0 - 37 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 2.492 ms | 3 - 8 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 0.966 ms | 0 - 21 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® SA7255P | 4.654 ms | 3 - 25 MB | NPU |
| XLSR | TFLITE | float | Qualcomm® SA8295P | 3.079 ms | 0 - 18 MB | NPU |
| XLSR | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.966 ms | 0 - 21 MB | NPU |
| XLSR | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.785 ms | 0 - 23 MB | NPU |
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.263 ms | 0 - 29 MB | NPU |
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.47 ms | 0 - 34 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 1.131 ms | 0 - 3 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.43 ms | 0 - 3 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.404 ms | 0 - 2 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® SA8775P | 0.607 ms | 0 - 23 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® SA8650P | 0.607 ms | 0 - 23 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® SA8255P | 0.607 ms | 0 - 23 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® QCS8450 | 0.47 ms | 0 - 34 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.539 ms | 0 - 3 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 1.621 ms | 0 - 18 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.418 ms | 0 - 20 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.215 ms | 0 - 21 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® SA7255P | 0.947 ms | 0 - 22 MB | NPU |
| XLSR | TFLITE | w8a8 | Qualcomm® SA8295P | 0.793 ms | 0 - 18 MB | NPU |
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 0.215 ms | 0 - 21 MB | NPU |
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.186 ms | 0 - 21 MB | NPU |
| XLSR | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.418 ms | 0 - 20 MB | NPU |
License
- The license for the original implementation of XLSR can be found here.
References
- Extremely Lightweight Quantization Robust Real-Time Single-Image Super Resolution for Mobile Devices
- 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.
