Hanwen Jiang

> The error is because cuda 11 no longer supports sm30 as target architecture. I think you can fix easily, by changing `sm_30` to `sm_60` in `dirt/csrc/CMakeLists.txt` line 41. Note...

Hi, thanks for the question. In training, we do not apply transformation augmentation (while the ObMan dataset has a small object translation coverage in 3D). So the model will not...

> > Hi, thanks for the question. > > In training, we do not apply transformation augmentation (while the ObMan dataset has a small object translation coverage in 3D). So...

> Thanks for your quick reply. > > Will you be able to help me understand the object distribution that you are using? Or if it depends on the ObMan...

Yes, you should scale the input object point cloud to roughly match the size of the hand

Hi, it seems your setting is correct. Have you evaluated the penetration depth of the ObMan GT?

There might be some problems with your GT. ObMan dataset doesn't provide the MANO parameters for each sample, do you directly use the hand vertices? And have you visualized your...

Directly using vertices is okay. Have you visualized your data? penetration of GT should be small.

Hi Fenggen, thanks for your interest in our work! Sorry, I missed the question last week. Yes, you can use [this function](https://github.com/UT-Austin-RPL/FORGE/blob/f0a30d19affd7fdd977604466c3b3e57f30756e9/scripts/kubric_validation.py#L15) for evaluation, which is used for validation in...

> > Hi, thanks for the question. > > In training, we do not apply transformation augmentation (while the ObMan dataset has a small object translation coverage in 3D). So...