Key not found and incorrect shape warnings when loading ckpt with load_checkpoint_with_missing_or_exsessive_keys.

[kohjakob]

When using def load_checkpoint_with_missing_or_exsessive_keys(cfg, model) to load unscene3d_DINO_CSC_self_train_3 into models.mask3d.Mask3D, i get the following warnings:

2025-03-31 12:41:31.577 | WARNING  | external.UnScene3D.utils.utils:load_checkpoint_with_missing_or_exsessive_keys:102 - Key not found, it will be initialized randomly: criterion.empty_weight
2025-03-31 12:41:31.741 | WARNING  | external.UnScene3D.utils.utils:load_checkpoint_with_missing_or_exsessive_keys:114 - incorrect shape backbone.conv0p1s1.kernel:torch.Size([125, 3, 32]) vs torch.Size([27, 3, 32])
2025-03-31 12:41:31.741 | WARNING  | external.UnScene3D.utils.utils:load_checkpoint_with_missing_or_exsessive_keys:114 - incorrect shape class_embed_head.weight:torch.Size([3, 128]) vs torch.Size([19, 128])
2025-03-31 12:41:31.741 | WARNING  | external.UnScene3D.utils.utils:load_checkpoint_with_missing_or_exsessive_keys:114 - incorrect shape class_embed_head.bias:torch.Size([3]) vs torch.Size([19])
2025-03-31 12:41:31.743 | WARNING  | external.UnScene3D.utils.utils:load_checkpoint_with_missing_or_exsessive_keys:126 - excessive key: criterion.empty_weight

Is this expected behavior or should i load the checkpoint differently?

I find the instances in the appended screenshot for scannets scene0000_00 when merging instance pairs with iou_matrix > 0.5 while propagating highest mask_confidence to merged instance and thresholding merged instances by mask_confidence > 0.9.