Verg-Avesta
commented
1 year ago Thanks for your attention.
I might not have clearly understood your meaning. Do you mean that you fine-tuned the pre-trained checkpoints provided in this issue and get the similar results, but you pre-trained a model from scratch and then fine-tuned it but get a bad result?
If that's the problem, I might have an answer. In the MAE pre-training stage, I didn't pre-train the model from scartch, instead, I used the checkpoints pre-trained on ImageNet by MAE in this issue. You can try whether this will give you a similar result.
Besides, I strongly recommend you to visualize the model performance pre-trained with MAE (i.e. after the first stage) and check whether the model has learned to reconstruct the masked images.
Xu3XiWang
Hello. Thank you for your great work.
I used the pre-trained model you provided and tried the second stage parameters provided in the readme.md and got similar results to those in the paper.
However, when I used the first stage parameters provided in the document, started training on the first stage using FSC-147, and then trained on the second stage with the parameters provided in the document, I ended up with the following results. MAE: 23.76, RMSE: 105.93
I think I have a problem with the first stage training. Is the first stage pre-training model in the paper obtained by using the parameters in the readme.md, and also is there anything else that needs to be modified?