[Feature] Support ScanRefer or nr3d/sr3d data set in Multi-View 3D Visual Grounding Task.

[LZ-CH]

What is the feature?

Hi, Thank you very much for your great work. Have you tried training and verifying this benchmark on scanrefer or nr3d/sr3d datasets? If so, can you publish the code that supports training and verification in other 3D-grounding datasets? This can provide great convenience for researcher to further improve and evaluate based on your benchmark in Multi-View 3D Visual Grounding Task.

Any other context?

No response

[Tai-Wang]

Thanks for your suggestion. It is also important to evaluate methods on conventional 3D visual grounding benchmarks. We will consider supporting this feature here or in Grounded 3D-LLM afterward.

[LZ-CH]

Thanks! Looking forward to your great work in the future!!!