The number of iterations in each of the training steps in your paper

[nicesonnday]

Hi, I wonder the number of iterations in your paper.

You said that you trained the R2L model in 600k iterations. (For ablation, 200k iterations)

But you didnt't mentioned that the number of iterations for pseudo data and real data training, respectively.

Thank you

[MingSun-Tse]

Hi @nicesonnday , 600k is for the pseudo data training. For the real data finetuning, it is pretty fast and up to specific scenes (on some scenes, it only takes ~1k iters or even fewer before overfitting). There is no one-for-all iteration number for the finetuning process. I would suggest trying it out based on the specific scene you're working on.