yolov3

This repository is an implementation of YOLOv3, an object detection model, maintained by Ultralytics. Object detection is the task of having a computer look at an image or video frame and identify what objects appear in it and where they are, drawing bounding boxes around each one. YOLO, which stands for You Only Look Once, is a family of models known for doing this quickly enough to work on video in real time.

YOLOv3 was a significant version in the YOLO series released in 2018, improving accuracy on small objects compared to earlier versions. This Ultralytics version reimplements it in PyTorch, a widely used Python library for building AI models, and adds tooling for training on custom datasets, running inference on images or video, and exporting the trained model to other formats including ONNX, CoreML, and TFLite. These export formats let you deploy the model on different platforms, including mobile devices.

To use it, you clone the repository, install the Python dependencies, and then run command-line scripts for training or detection. Training requires providing labeled images where someone has already drawn the bounding boxes so the model can learn from them. After training, you run a detection script pointing it at an image, video file, or camera feed, and it outputs the image with boxes drawn around detected objects.

The README notes that Ultralytics now recommends their newer model family, YOLO11, for new projects, as it is more capable. This YOLOv3 repository remains available for research, compatibility, and learning purposes.

The repository includes a Docker image for running it in a container, integration with Google Colab and Kaggle notebooks for trying it without a local setup, and export support for deployment on edge hardware. The code is licensed under the AGPL-3.0 license for open-source use, with a commercial enterprise license available separately.