README.md

---
library_name: pytorch
license: other
tags:
- bu_auto
- real_time
- android
pipeline_tag: object-detection

---

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

# Yolo-R: Optimized for Qualcomm Devices

YoloR is a machine learning model that predicts bounding boxes and classes of objects in an image.

This is based on the implementation of Yolo-R found [here](https://github.com/WongKinYiu/yolor).
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/yolor) 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
Due to licensing restrictions, we cannot distribute pre-exported model assets for this model.
Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/yolor) 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

See our repository for [Yolo-R on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/yolor) for usage instructions.


## Model Details

**Model Type:** Model_use_case.object_detection

**Model Stats:**
- Model checkpoint: yolor_p6
- Input resolution: 640x640
- Number of parameters: 4.68M
- Model size (float): 17.9 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Yolo-R | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 24.445 ms | 6 - 316 MB | NPU
| Yolo-R | ONNX | float | Snapdragon® 8 Elite Mobile | 25.096 ms | 2 - 232 MB | NPU
| Yolo-R | ONNX | float | Snapdragon® X2 Elite | 24.767 ms | 75 - 75 MB | NPU
| Yolo-R | ONNX | float | Snapdragon® X Elite | 38.214 ms | 74 - 74 MB | NPU
| Yolo-R | ONNX | float | Snapdragon® X Elite | 38.214 ms | 74 - 74 MB | NPU
| Yolo-R | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 28.964 ms | 0 - 341 MB | NPU
| Yolo-R | ONNX | float | Qualcomm® QCS8550 (Proxy) | 38.039 ms | 0 - 78 MB | NPU
| Yolo-R | ONNX | float | Qualcomm® QCS9075 | 52.16 ms | 5 - 12 MB | NPU
| Yolo-R | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 25.096 ms | 2 - 232 MB | NPU
| Yolo-R | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 18.066 ms | 3 - 452 MB | NPU
| Yolo-R | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 17.404 ms | 1 - 366 MB | NPU
| Yolo-R | ONNX | w8a16 | Snapdragon® X2 Elite | 18.504 ms | 41 - 41 MB | NPU
| Yolo-R | ONNX | w8a16 | Snapdragon® X Elite | 30.398 ms | 40 - 40 MB | NPU
| Yolo-R | ONNX | w8a16 | Snapdragon® X Elite | 30.398 ms | 40 - 40 MB | NPU
| Yolo-R | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 20.837 ms | 0 - 492 MB | NPU
| Yolo-R | ONNX | w8a16 | Qualcomm® QCS6490 | 2272.484 ms | 129 - 135 MB | CPU
| Yolo-R | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 29.064 ms | 0 - 49 MB | NPU
| Yolo-R | ONNX | w8a16 | Qualcomm® QCS9075 | 29.698 ms | 1 - 6 MB | NPU
| Yolo-R | ONNX | w8a16 | Qualcomm® QCM6690 | 1177.534 ms | 91 - 103 MB | CPU
| Yolo-R | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 17.404 ms | 1 - 366 MB | NPU
| Yolo-R | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1120.199 ms | 136 - 148 MB | CPU
| Yolo-R | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1120.199 ms | 136 - 148 MB | CPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 7.916 ms | 2 - 305 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 10.167 ms | 2 - 304 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 8.914 ms | 2 - 2 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® X Elite | 20.753 ms | 2 - 2 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® X Elite | 20.753 ms | 2 - 2 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 13.32 ms | 2 - 356 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 78.992 ms | 1 - 6 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 39.816 ms | 2 - 291 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 19.887 ms | 2 - 272 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® SA8775P | 19.89 ms | 2 - 291 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® SA8775P | 19.89 ms | 2 - 291 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® SA8775P | 19.89 ms | 2 - 291 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 20.74 ms | 1 - 6 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 214.15 ms | 2 - 400 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 28.632 ms | 2 - 358 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® SA7255P | 39.816 ms | 2 - 291 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Qualcomm® SA8295P | 25.586 ms | 0 - 293 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 10.167 ms | 2 - 304 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 26.722 ms | 2 - 318 MB | NPU
| Yolo-R | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 26.722 ms | 2 - 318 MB | NPU

## License
* The license for the original implementation of Yolo-R can be found
  [here](https://github.com/WongKinYiu/yolor/blob/main/LICENSE).

## References
* [You Only Learn One Representation: Unified Network for Multiple Tasks](https://arxiv.org/abs/2105.04206)
* [Source Model Implementation](https://github.com/WongKinYiu/yolor)

## 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).