problem when trying to finetune a model, using resnet50 backbone

[saeidrazavi]

when i try to fine-tune pre_trained model (using full ckpt) with a different dataset using following command :

!python "main_dino.py" --arch resnet50 --batch_size_per_gpu 32 --epochs=20 --data_path "/content/dataset/train" --output_dir "/content/model_checkpts"

i got this error :

size mismatch for module.head.mlp.0.weight: copying a param with shape torch.Size([4096, 2048]) from checkpoint, the shape in current model is torch.Size([2048, 2048]).
size mismatch for module.head.mlp.0.bias: copying a param with shape torch.Size([4096]) from checkpoint, the shape in current model is torch.Size([2048]).

can someone explain me the problem?
i think the problem is that full_ckpt file, does not contain full weights of mlp heads and it's kinda incomplete

[Badar-e-Alam]

Because you have a checkpoint in output_dir which will be loaded by utils.restart_from_checkpoint , another hint is you use the arch as resent50 it's also have the Dino_head and MultiCropWrapper warp around look line 133.