Congratulations on your fatasitic works. To apply your framework on my own dataset, which is to complete point cloud obtained by a depth camera with 114 categories, I have to handle with sub point clouds with different unknown poses.
However, I noticed that there is no transforms similar to RandomRotation things in your dataloader. The fact is when I tried to train a net that fits the data all with default poses, the results is great. But when I added RandomRotation in the dataloader, the inference outcomes are bad.
So please tell me if your work could resolve such cases and whether the datasets you used include such rotation itself? If possible, please offer me suggestions to do adjustment.
SLDRMK
commented
3 months ago
Dear authers,
Congratulations on your fatasitic works. To apply your framework on my own dataset, which is to complete point cloud obtained by a depth camera with 114 categories, I have to handle with sub point clouds with different unknown poses.
However, I noticed that there is no transforms similar to RandomRotation things in your dataloader. The fact is when I tried to train a net that fits the data all with default poses, the results is great. But when I added RandomRotation in the dataloader, the inference outcomes are bad.
So please tell me if your work could resolve such cases and whether the datasets you used include such rotation itself? If possible, please offer me suggestions to do adjustment.