search

valeoai / xmuda

Cross-Modal Unsupervised Domain Adaptationfor 3D Semantic Segmentation
Other
192 stars 36 forks source link

AssertionError when generating pseudo labels #13

Closed YukangWang closed 3 years ago

YukangWang commented 3 years ago

Hi, I'm currently using my own dataset as the target. Since this dataset is unlabeled, I modify the related files and finally run train_xmuda.py sucessfully. But when I try to generate pseudo labels, I encounter an AssertionError at: https://github.com/valeoai/xmuda/blob/8b39f8f77f048cd4c086323dd89fbf9710335d35/xmuda/data/utils/validate.py#L60 I wonder why and what would happen if I ignore this error. Thanks in advance.

maxjaritz commented 3 years ago

Hi, SparseConvNet has maximum dimensions for the input coordinate grid, defined by the full_scale parameter. If you have points outside of full_scale, they will be discarded in the dataloader. When you predict the pseudo labels, it will check if you have valid labels for all points. Thus, if some points were discarded in the dataloader, because they were outside the full_scale, it throws an error. One easy way to solve it is to set the missing labels to ignore (-100).

YukangWang commented 3 years ago

Got it, thanks!