Open dongxinfeng1 opened 1 month ago
Hi, You should first extract the image features using the provided scripts. After that, you should run the pre-training scripts to get the .pt file. Then, you can choose one .pt file to initialize your fine-tuning stage.
I run the checkpoint you provide "ckpts/r2r_best" and it works well. In the code your provide, I can train the model and get the ".pt" file for checkpoint which cannot be used for fine-tuning and testing directly. Do you know how can I get my own checkpoint for fine-tuning and testing in R2R? Hope for your reply and thanks a lot.