search

GuiyuZhao / VRHCF

[ICME 2024] VRHCF: Cross-Source Point Cloud Registration via Voxel Representation and Hierarchical Correspondence Filtering
MIT License
22 stars 3 forks source link

Question #2

Open JiaPeng-W opened 6 months ago

JiaPeng-W commented 6 months ago

Hello author, thank you for sharing your code! I noticed that the paper mentions using FPS downsampling and KNN search on the original point cloud, but I couldn't find the corresponding parts in the provided code. Do you know where these parts are located? Thank you.

GuiyuZhao commented 6 months ago

Hi. For FPS downsampling, we get the keypoints from the keypoint file by preprocessing the original point cloud (line 31 in preparation.py), without resampling the point. For KNN search, we use the query_radius function from the sklearn library (in line 33 of preparation.py). It is worth noting that in order to train quickly, local patch point clouds are obtained by processing in advance during training, so downsampling and KNN are not required again.