Open supgy opened 7 months ago
I am not familiar with partial point cloud generation. Usually, this should be done by projecting a 3D mesh to a depth map given a camera pose, then back project this depth map to 3D to create self-occlusion-based incompleteness. To this end, you can have a look on:
I am not familiar with partial point cloud generation. Usually, this should be done by projecting a 3D mesh to a depth map given a camera pose, then back project this depth map to 3D to create self-occlusion-based incompleteness. To this end, you can have a look on:
Hi,I mean that, I notice that each complete point cloud corresponds to eight partial point clouds in EPN-3D database,I want to know how partial point cloud generate,thanks : ) for example, a completion pcd and its coresspond partial pcd as show below:
The data generation process was detailed in 3D-EPN paper, Sec. 4 https://arxiv.org/pdf/1612.00101.pdf We used the point cloud format sampled by Wen et al. (Cycle4Completion)
Cycle4Completion
Got that, thanks for your response!
Hi,thank you for answering everyone's questions. I want to know how to generate partial data based on completion data?I want to train a model from scratch using my own data. Or how do I build my own dataset(completion point cloud、partial point cloud and labels?)