How to do inference on customized data?

[Cao-Cong0]

For example I have a obj file, I need to convert it to glb file and then use blender to render rgb and depth, right? What else do I need to input?

And can it work on the 3d model without texture?

[yhyang-myron]

Hi, thanks for your interest in our work!

1. You can directly render images and train MLPs using our codes for OBJ files.
2. It can work on the 3D model without texture. You can set the color value (after rgb augmentation) to (0, 0, 0) for 3d models without texture during the training stage.

[Batwho]

Hi, thanks for your interest in our work!

1. You can directly render images and train MLPs using our codes for OBJ files.

2. It can work on the 3D model without texture. You can set the color value (after rgb augmentation) to (0, 0, 0) for 3d models without texture during the training stage.

For the second points, could you please point out where is the code to set this? Also, I test SAMPart3D on some objects in ModelNet40 and the segmentation sometimes are pretty bad. Do you think that might be because of the lack of texture?

[yhyang-myron]

The color value can be set to (0, 0, 0) after transform.

Could you please check the SAM segmentation results or whether the mapping between 2D pixels and 3D points is right?