How to use a pre-trained PTv3-obj model to obtain 3D features?

[llcc343]

Hi,
I already have a custom mesh that has been normalized to the range [-1, 1], and I have sampled points and their corresponding normals. I have wrapped them into a dictionary and input them into the model built with build_model, but the forward pass is not working. In this case, how can I use it correctly?

[yhyang-myron]

Hi, thanks for your interest in our work!
Could you please clarify what you mean by "forward pass is not working"?
You can obtain the 3D features from pre-trained PTv3-object like this.

[llcc343]

Thanks for your kind repply. I use the model configuration from

SAMPart3D/configs/sampart3d/sampart3d-trainmlp-render16views.py

Line 19 in 6f889d7

model = dict(

with the build_model() method. And then I put the sampled points normals, coords and offsets(calculated using transform.py) into a dict , and pass the dict to the model forward, but offset2batch in the Point(Dict) crashed.