Samples in val dataset present inside train dataset

[kbali1297]

When I extracted the dataset pkl file I found that most validation samples are also present in the training dataset. The shape ids in the validation set are exactly present inside the training set and the meshes are almost the same. It would be good if these mesh samples were removed from training set for testing the network's generalization better.

[nihalsid]

Hi, thanks for your pointing out this issue! Good catch. Agree that the val set shouldnt have been mixed with train, it seemed to have leaked in because of the different decimation prefix in the name. Feel free to make a proper val set from the train names by doing the split with the cateogory + shape id instead for better tracking of training. That being said, the purpose of the val set is only tracking training runs, and not testing.