README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: image-to-image

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/xlsr/web-assets/model_demo.png)

# 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](https://github.com/quic/aimet-model-zoo/tree/develop/aimet_zoo_torch/xlsr).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/xlsr) 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 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/xlsr/releases/v0.54.0/xlsr-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/xlsr/releases/v0.54.0/xlsr-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/xlsr/releases/v0.54.0/xlsr-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/xlsr/releases/v0.54.0/xlsr-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/xlsr/releases/v0.54.0/xlsr-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/xlsr/releases/v0.54.0/xlsr-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[XLSR on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/xlsr)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/xlsr) 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](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/xlsr) for usage instructions.

## Model Details

**Model Type:** Model_use_case.super_resolution

**Model Stats:**
- Model checkpoint: xlsr_3x_checkpoint
- Input resolution: 128x128
- Number of parameters: 28.0K
- Model size (float): 115 KB
- Model size (w8a8): 45.6 KB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| XLSR | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.423 ms | 0 - 22 MB | NPU
| XLSR | ONNX | float | Snapdragon® 8 Elite Mobile | 0.506 ms | 0 - 18 MB | NPU
| XLSR | ONNX | float | Snapdragon® X2 Elite | 0.4 ms | 6 - 6 MB | NPU
| XLSR | ONNX | float | Snapdragon® X Elite | 1.061 ms | 9 - 9 MB | NPU
| XLSR | ONNX | float | Snapdragon® X Elite | 1.061 ms | 9 - 9 MB | NPU
| XLSR | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.65 ms | 0 - 29 MB | NPU
| XLSR | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.935 ms | 0 - 2 MB | NPU
| XLSR | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.506 ms | 0 - 18 MB | NPU
| XLSR | ONNX | float | Qualcomm® QCS9075 | 1.233 ms | 8 - 11 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.245 ms | 0 - 23 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 0.264 ms | 0 - 24 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® X2 Elite | 0.227 ms | 1 - 1 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® X Elite | 0.651 ms | 3 - 3 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® X Elite | 0.651 ms | 3 - 3 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.365 ms | 0 - 30 MB | NPU
| XLSR | ONNX | w8a8 | Qualcomm® QCS6490 | 12.168 ms | 19 - 21 MB | CPU
| XLSR | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.549 ms | 0 - 67 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 6.443 ms | 20 - 26 MB | CPU
| XLSR | ONNX | w8a8 | Qualcomm® QCM6690 | 8.508 ms | 18 - 25 MB | CPU
| XLSR | ONNX | w8a8 | Qualcomm® QCS9075 | 0.702 ms | 0 - 3 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.264 ms | 0 - 24 MB | NPU
| XLSR | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 6.443 ms | 20 - 26 MB | CPU
| XLSR | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.337 ms | 0 - 24 MB | NPU
| XLSR | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.357 ms | 0 - 25 MB | NPU
| XLSR | QNN_DLC | float | Snapdragon® X2 Elite | 0.411 ms | 0 - 0 MB | NPU
| XLSR | QNN_DLC | float | Snapdragon® X Elite | 0.895 ms | 0 - 0 MB | NPU
| XLSR | QNN_DLC | float | Snapdragon® X Elite | 0.895 ms | 0 - 0 MB | NPU
| XLSR | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.461 ms | 0 - 29 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 0.763 ms | 0 - 1 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® SA8775P | 1.144 ms | 0 - 22 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® SA8775P | 1.144 ms | 0 - 22 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® SA8775P | 1.144 ms | 0 - 22 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® SA7255P | 2.205 ms | 0 - 21 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 1.043 ms | 0 - 31 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® SA8295P | 1.403 ms | 0 - 18 MB | NPU
| XLSR | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.357 ms | 0 - 25 MB | NPU
| XLSR | QNN_DLC | float | Qualcomm® QCS9075 | 1.11 ms | 0 - 5 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.18 ms | 0 - 25 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 0.211 ms | 0 - 26 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.272 ms | 0 - 0 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.493 ms | 0 - 0 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.493 ms | 0 - 0 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.248 ms | 0 - 30 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 1.399 ms | 0 - 2 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.393 ms | 0 - 1 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.579 ms | 0 - 22 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.579 ms | 0 - 22 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.579 ms | 0 - 22 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.41 ms | 0 - 20 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 1.602 ms | 0 - 18 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.554 ms | 0 - 2 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.211 ms | 0 - 26 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.468 ms | 0 - 31 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA7255P | 0.91 ms | 0 - 21 MB | NPU
| XLSR | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.78 ms | 0 - 18 MB | NPU
| XLSR | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.41 ms | 0 - 20 MB | NPU
| XLSR | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.798 ms | 0 - 25 MB | NPU
| XLSR | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.987 ms | 0 - 27 MB | NPU
| XLSR | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.414 ms | 0 - 31 MB | NPU
| XLSR | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 2.222 ms | 0 - 2 MB | NPU
| XLSR | TFLITE | float | Qualcomm® SA8775P | 2.662 ms | 0 - 23 MB | NPU
| XLSR | TFLITE | float | Qualcomm® SA8775P | 2.662 ms | 0 - 23 MB | NPU
| XLSR | TFLITE | float | Qualcomm® SA8775P | 2.662 ms | 0 - 23 MB | NPU
| XLSR | TFLITE | float | Qualcomm® SA7255P | 4.713 ms | 3 - 25 MB | NPU
| XLSR | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 2.515 ms | 0 - 32 MB | NPU
| XLSR | TFLITE | float | Qualcomm® SA8295P | 3.14 ms | 0 - 18 MB | NPU
| XLSR | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.987 ms | 0 - 27 MB | NPU
| XLSR | TFLITE | float | Qualcomm® QCS9075 | 2.596 ms | 3 - 8 MB | NPU
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.18 ms | 0 - 24 MB | NPU
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 0.227 ms | 0 - 21 MB | NPU
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.265 ms | 0 - 29 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® QCS6490 | 1.083 ms | 0 - 3 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.398 ms | 0 - 2 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® SA8775P | 0.598 ms | 0 - 23 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® SA8775P | 0.598 ms | 0 - 23 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® SA8775P | 0.598 ms | 0 - 23 MB | NPU
| XLSR | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.419 ms | 0 - 19 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® QCM6690 | 1.623 ms | 0 - 18 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.602 ms | 0 - 3 MB | NPU
| XLSR | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.227 ms | 0 - 21 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.476 ms | 0 - 30 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® SA7255P | 0.962 ms | 0 - 21 MB | NPU
| XLSR | TFLITE | w8a8 | Qualcomm® SA8295P | 0.81 ms | 0 - 18 MB | NPU
| XLSR | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.419 ms | 0 - 19 MB | NPU

## License
* The license for the original implementation of XLSR can be found
  [here](https://github.com/quic/aimet-model-zoo/blob/develop/LICENSE.md).

## References
* [Extremely Lightweight Quantization Robust Real-Time Single-Image Super Resolution for Mobile Devices](https://arxiv.org/abs/2105.10288)
* [Source Model Implementation](https://github.com/quic/aimet-model-zoo/tree/develop/aimet_zoo_torch/xlsr)

## 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).