About the "3D + SAM" method mentioned in the paper:

[XXXCARREY]

I notice that the best results on Class-agnostic evaluation on ScanNet200 are based on the "3D + SAM" model. Does this SAM refer to the original SAM? I checked the code, and it seems that the segmenter2d does not include an option for the original SAM; it only has options like Grounded-SAM, etc. Could you please tell me how to reproduce the highest-performing method mentioned in the paper? Thank you very much.

[PhucNDA]

Hi @XXXCARREY,

Yes, It is the AutoMaskGenSAM from SegmentAnything-HQ. We haven't included it in the current version of the source code. Thanks for letting us know. We are currently supporting these types of segmenter2D (https://github.com/VinAIResearch/Open3DIS/blob/main/configs/scannet200.yaml#L19C2-L19C3). We will add it, soon.

Best,
PhucNDA.