YOLOv10

YOLOv10 is a state-of-the-art, end-to-end object detection model, pioneering NMS-free training with consistent dual assignments for superior real-time speed and accuracy.

This is the official PyTorch implementation of YOLOv10, developed by researchers at Tsinghua University. The model significantly advances the YOLO series by introducing a holistic efficiency-accuracy driven design and, critically, consistent dual assignments for NMS-free (Non-Maximum Suppression-free) training. Eliminating NMS drastically reduces post-processing overhead, ensuring lower end-to-end latency. For example, the YOLOv10-S model is 1.8x faster than RT-DETR-R18 at similar Average Precision (AP) on the COCO dataset, and YOLOv10-B achieves the same performance as YOLOv9-C with 46% less latency and 25% fewer parameters. It sets a new benchmark for efficient, real-time object detection.

https://github.com/THU-MIG/YOLOv10