This project re-implements the Faster R-CNN object detection algorithm in PyTorch. It provides a minimal, readable codebase (~2000 lines) that replicates the original paper’s performance (0.699 mAP on VOC), achieving comparable speed at 6fps training and 14fps testing on a TITAN XP GPU. The implementation uses torchvision’s built-in ROI pooling and NMS operations, removing the need for custom CUDA compilation.