meituan/yolov6

YOLOv6 is an object detection system built by Meituan, the large Chinese technology company. Object detection means feeding an image to the system and getting back a list of what objects are in the image and where they are located, each marked with a bounding box. This kind of technology is used in manufacturing quality control, security cameras, autonomous vehicles, and similar industrial settings. The framework comes in several model sizes labeled N (nano), S (small), M (medium), and L (large), plus larger P6 variants designed for higher-resolution inputs. Smaller models run faster but are less accurate, larger models are slower but detect objects with more precision. The README includes benchmark tables showing the trade-off: the nano model processes roughly 779 frames per second on an Nvidia T4 GPU, while the large model runs at 98 frames per second but achieves noticeably better accuracy scores on the standard COCO benchmark dataset. Beyond the core detection models, the project has branched into specialized variants. YOLOv6-Face targets face detection specifically. YOLOv6-Seg adds instance segmentation, which means identifying not just where objects are but outlining their exact shape. YOLOv6Lite is a stripped-down version designed to run on mobile devices or CPUs without a dedicated GPU. The project is built with PyTorch and includes tools for training new models, running inference on images or video, exporting to deployment formats like ONNX and TensorRT, and applying quantization to shrink model size further. Tutorials are available on Google Colab and Kaggle for people who want to experiment without setting up a local environment. The repository includes two research papers that describe the architecture and design choices behind different versions of the framework. It is released as open source under the GPL-3.0 license.