invalid argument 5: k not in range for dimension at /pytorch/aten/src/THC/generic/THCTensorTopK.cu:26 about overlappredator HOT 4 CLOSED

hi,

The bug occurs when you have less than 10 superpoints in the bottleneck(then the DGCNN part fails). Please consider adjusting k here: https://github.com/ShengyuH/OverlapPredator/blob/9772bb65fc06b9f576054620a3a2da7a49d4ace7/configs/test/modelnet.yaml#L32

We use try/catch scheme to skip these pairs in our experiments.

from overlappredator.

hi,

The bug occurs when you have less than 10 superpoints in the bottleneck(then the DGCNN part fails). Please consider adjusting k here:

https://github.com/ShengyuH/OverlapPredator/blob/9772bb65fc06b9f576054620a3a2da7a49d4ace7/configs/test/modelnet.yaml#L32

We use try/catch scheme to skip these pairs in our experiments.

Thanks for your reply. I change dfcnn_k to 1, but it is useless . I fixed that by change first_subsampling_dl to 0.01. The bug maybe caused by subsampling.

from overlappredator.

hi,

Please be really careful when changing first_subsampling_dl value. This determines the effective receptive field and thus the final performance.

from overlappredator.

Please reopen this issue if your find further questions

from overlappredator.