This report introduces YOLOX, a high-performance anchor-free object detection model that builds upon the YOLO series. The authors present several improvements, including a decoupled head, anchor-free detection, and advanced label assignment strategies like SimOTA. These enhancements enable YOLOX to achieve state-of-the-art results across a wide range of models, from small to large. Specifically: - **YOLOX-Nano**: Achieves 25.3% AP on COCO, surpassing NanoDet by 1.8% AP. - **YOLOv3**: Boosted to 47.3% AP on COCO, outperforming the current best practice by 3.0% AP. - **YOLOX-L**: Achieves 50.0% AP on COCO at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP. The report also highlights the use of advanced data augmentation techniques, such as Mosaic and MixUp, and the benefits of an anchor-free mechanism. YOLOX is designed to be flexible and efficient, with deploy versions supported by ONNX, TensorRT, NCNN, and Openvino. Additionally, the authors won the 1st Place on the Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. The report aims to provide valuable insights and practical guidance for developers and researchers in the field of object detection.This report introduces YOLOX, a high-performance anchor-free object detection model that builds upon the YOLO series. The authors present several improvements, including a decoupled head, anchor-free detection, and advanced label assignment strategies like SimOTA. These enhancements enable YOLOX to achieve state-of-the-art results across a wide range of models, from small to large. Specifically: - **YOLOX-Nano**: Achieves 25.3% AP on COCO, surpassing NanoDet by 1.8% AP. - **YOLOv3**: Boosted to 47.3% AP on COCO, outperforming the current best practice by 3.0% AP. - **YOLOX-L**: Achieves 50.0% AP on COCO at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP. The report also highlights the use of advanced data augmentation techniques, such as Mosaic and MixUp, and the benefits of an anchor-free mechanism. YOLOX is designed to be flexible and efficient, with deploy versions supported by ONNX, TensorRT, NCNN, and Openvino. Additionally, the authors won the 1st Place on the Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. The report aims to provide valuable insights and practical guidance for developers and researchers in the field of object detection.