keras-yolo2

Why we included this project

Training a custom object detector usually means wrestling with framework plumbing, but this repo gives you a working YOLOv2 implementation in Keras with a TensorFlow backend, ready to train on your own VOC-format annotations and image folders. You can swap lightweight backbones like MobileNet, SqueezeNet, InceptionV3, or ResNet50, which is handy when you want to trade a little accuracy for faster inference on less powerful hardware. The bundled examples for raccoon, kangaroo, hand, and red blood cell detection double as starting templates, and there is even a raccoon detector demo that runs entirely in the browser. For people who want to see how the training pieces fit together rather than run a black box, the walkthrough of data preparation and training steps is a practical way in.