lilanxiao/Rotated_IoU
A PyTorch library with CUDA extensions for computing differentiable IoU and loss functions on oriented bounding boxes in 2D and 3D.

Velocity · 7d
+0.2
★ / day
Trend
→steady
star history
This repository provides CUDA-accelerated functions to calculate the intersection area of rotated rectangles using PyTorch, enabling backpropagation through IoU computation. It implements GIoU and DIoU loss variants for oriented bounding boxes in both 2D and 3D spaces. The library includes a training demo that generates synthetic box data and validates the gradients for object detection loss functions.