CuiRuikai / Partial2Complete

[ICCV 2023] P2C: Self-Supervised Point Cloud Completion from Single Partial Clouds
MIT License
159 stars 9 forks source link

Tensors of type TensorImpl do not have sizes #25

Open Xxxkkkkx opened 2 months ago

Xxxkkkkx commented 2 months ago

During the training process, I keep encountering the following error. I have checked the format of xyz1 and xyz2 inputs to ChamferFunction, which are (32, 640, 3) and (32, 1280, 3) respectively, and they seem to be correct. I would like to consult on possible solutions. Looking forward to your response. 错误

CuiRuikai commented 2 months ago

I did not encounter such a problem. After some research on this issues, I have two things to confirm: 1) are you using sparse tensor? 2) can you try making all tensor continuous (.continuous() in PyTorch) before sending to the chamfer function.

CuiRuikai commented 2 months ago

PyTorch3D also provides an implementation of Chamfer distance, which might be more robust than the one that I used. You could try this version. https://pytorch3d.readthedocs.io/en/latest/modules/loss.html

Please let me know if the above solutions are not able to resolve this issues