search

zxhuang1698 / ZeroShape

Code repository for "ZeroShape: Regression-based Zero-shot Shape Reconstruction".
https://zixuanh.com/projects/zeroshape.html
109 stars 10 forks source link

Some questions about gt_sample_sdf #5

Closed QiuJunning closed 1 day ago

QiuJunning commented 1 week ago

Hello, I think this work is very interesting and great!!!!! I have a question about the dataset labeling, I hope you can answer it. I noticed that the true value of occupancy is obtained by reading sample_sdf and subtracting 0.003. I want to know if sample_sdf is the definition of SDF? Why do we need to subtract 0.003? And how can I get this sample_sdf value of a model? Looking forward to your answer! Thank you!

zxhuang1698 commented 1 week ago

Thanks for your interest in our work. This offset is related to the data preparation code we are using. We first remove the interior surfaces with virtual scanning (https://github.com/davidstutz/mesh-fusion) and then convert it to SDF query and GT following https://github.com/rehg-lab/3DShapeGen/blob/master/SDFNet/README.md.

QiuJunning commented 1 week ago

I can use the above process to remove the internal surface of my object model and then get the SDF label. After executing according to these two repos, I still need to keep the setting of minus 0.003. I don’t know if I understand it correctly?

zxhuang1698 commented 1 week ago

Yes that is correct.