Trained model for PartNet

[TangWaiLun-Brian]

Thanks for sharing the source code. Just one thing to clarify, I want to ask whether the ShapeFormer model is trained separately on different datasets in the Shape completion results section of your research paper. If yes, may I kindly ask for the sharing of the pre-trained model on PartNet. Thanks a lot!

[lucasbrynte]

I had the same question, and was able to infer from Section A.3 in the supplementary material that 3 ShapeFormer models seem to have been trained: one on ShapeNet, one on PartNet, and a third one on D-FAUST.

My guess would be that along with each of these ShapeFormer models, a corresponding VQDIF model is trained on the same data, but regarding this I am very uncertain. Perhaps @QhelDIV could shed some light on this..?

As a first step towards sharing additional pre-trained models as @TangWaiLun-Brian requested, would it be possible to publish additional training/testing configuration .yaml files under configs/, such that one could train not only the ShapeNet model, but also the PartNet / D-FAUST models?

Best Regards, Lucas