| license: other | |
| license_name: upstream-terms-by-artifact | |
| license_link: https://huggingface.co/HanzoHuang/YOLOX-RKNN2/blob/main/LICENSES.md | |
| pipeline_tag: object-detection | |
| tags: | |
| - rknn | |
| - rknn2 | |
| - rockchip | |
| - rk3576 | |
| - rk3588 | |
| - yolox | |
| # YOLOX-RKNN2 | |
| > **Looking for Ultralytics RKNN exports?** Browse the [Ultralytics exports collection](https://huggingface.co/collections/HanzoHuang/ultralytics-rknn-exports-6af1d088ba04ccf2dcbf81c2). | |
| Rockchip Model Zoo conversions of YOLOX Nano for **RK3576** and **RK3588**, with FP16 and INT8 variants. | |
| ## Choose a model | |
| For Toolkit 2.5.0, select a directory under `models/v2.5.0/detect/yolox_nano/<chip>/<precision>/` and download `model.rknn` and `metadata.yaml` together. The [manifest](manifest.json) records model hashes. | |
| - [Toolkit 2.5.0 builds](tree/main/models/v2.5.0) use RKNN Toolkit2 2.5.0. Conversion receipts identify the original ONNX model and converter. | |
| - [Toolkit 2.3.2 builds](tree/main/models/v2.3.2) were tested on their target board. Each receipt identifies its source graph and compiler. | |
| ## Conversion and license | |
| The [original YOLOX Nano checkpoint](https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_nano.pth) is included in [source](tree/main/source). It was exported to ONNX with the [Rockchip YOLOX exporter](https://github.com/airockchip/YOLOX/tree/fc69c1e2ccb30fce7e18250bddc13a68090dca09) and compiled with the [Model Zoo YOLOX converter](https://github.com/airockchip/rknn_model_zoo/tree/837db8e992815ff8647eadb709f07067f4bc3a71/examples/yolox) and RKNN Toolkit2 2.5.0. The INT8 variant was quantized during conversion. | |
| YOLOX is from [Megvii-BaseDetection](https://github.com/Megvii-BaseDetection/YOLOX). See [LICENSE](LICENSE) and [source and license notices](LICENSES.md) for the applicable terms. | |