size mismatch for model.head.feature_mapping.weight

[Bruno-Lages]

Hi, I'm trying to make some predictions in the mini dataset described in the Artic readme using the checkpoint disponibilized, but I'm facing some issues related to the checkpoint mismatch.

When running the extracct_predict function, as described in the artic repo
python scripts_method/extract_predicts.py --setup p1 --method transformer_sf --load_ckpt /app/final_p1_504802f70.ckpt --run_on val --extraction_mode eval_pose

I got the error:

size mismatch for model.head.feature_mapping.weight: copying a param with shape torch.Size([512, 1536]) from checkpoint, the shape in current model is torch.Size([512, 2048]).

Also, if possible, could you tell if it's necessary to do extra steps for the prediction?

[kabouzeid]

Sadly I can't help. I'm not very familiar with the codebase anymore at this point :/