Part-level pose estimation and texture-less mesh performance?

[1270645409]

Hi, I have two quick questions about FoundationPose:

• Part-level pose: Can it work when the CAD model is only a part (e.g., microwave handle), while the input RGB-D shows the full object? I can provide a mask and initial pose for the part. Since training uses full objects, does it still generalize? Any tricks needed?
• Texture-less meshes: How robust is it on low-texture or texture-less CAD models? Does it rely heavily on RGB appearance, or can geometry/depth cues compensate?

Thanks!