Question about pretraining-based methods.

[zixiangro]

Whether corruptions that are part of ModelNet-C and ShapeNet-C are allowed in the pretraining phase?

As we know these methods (e.g. OcCo, Point-BERT & Point-MAE) all use similar corruptions like drop local in unlabeled data.

[ldkong1205]

Hi @Xxiaoluohao, thanks for reaching us!

For your question, the common data augmentation techniques like drop local are allowed during the unsupervised pertaining phase. However, they are not permitted throughout the supervised training phase.

We do this to ensure that the improvements come from the model design itself, as our goal is to explore novel architectures that can improve recognition robustness.

Thanks for your question, it's a good one!