Training datasets for v2.1 version

[ZCMax]

Thanks for your great work! May I know whether you have done any dataset updation for the version 2.1?

[ZzZZCHS]

For evaluating on datasets such as Multi3DRefer and SQA3D, we add them (train split) into the training data. And in v2.1, we train the model in one single joint-training stage, so we delete some unnecessary data for object-level/scene-level alignment proposed in our paper. There leaves a large space for exploring to include more high-quality datasets during training.

[ZCMax]

So you remove the generated datasets in v2.1? only using the existing human-annotated 3d-vl datasets?

[ZzZZCHS]

Yes.