The official repository of the ICLR 2025 paper "GPS: A Probabilistic Distributional Similarity with Gumbel Priors for Set-to-Set Matching"
We update the demo code of our GPS
For implementation details, just follow any CD loss based point cloud completion model's code(Like CP-Net)
Here are some key items must to know: 1.We provide the pure python version of GPS 2.For the CUDA version like Pytorch Chamfer Distance You can add line 64 to line 90 in utils.py to Vanilla CD(the return distance in thier CD loss) for usage
