cszn / KAIR

Image Restoration Toolbox (PyTorch). Training and testing codes for DPIR, USRNet, DnCNN, FFDNet, SRMD, DPSR, BSRGAN, SwinIR

Home Page:https://cszn.github.io/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How to train RVRT with my own dataset?

skerit opened this issue · comments

I've got my own dataset of low-quality & corresponding high-quality frames as PNG files.
But I can't figure out how to actually train RVRT on them.
I've tried the create_lmdb script, but I just don't get it.

commented

Hello, can RVRT be trainde with multi GPU? I tried train RVRT using DDP/DP, but failed on deform_attn

I've got my own dataset of low-quality & corresponding high-quality frames as PNG files. But I can't figure out how to actually train RVRT on them. I've tried the create_lmdb script, but I just don't get it.

I met the same problem. I tried the create_lmdb script, but it seems that the structure of the GoPro (the same as other datasets) is not like what it is in the create_lmdb script.

I've got my own dataset of low-quality & corresponding high-quality frames as PNG files. But I can't figure out how to actually train RVRT on them. I've tried the create_lmdb script, but I just don't get it.

I met the same problem. I tried the create_lmdb script, but it seems that the structure of the GoPro (the same as other datasets) is not like what it is in the create_lmdb script.

Hello, I met a problem. I want to train SwinIR with my own dataset, But it is rectangular. I don't know where to change the size. do you know how to solve it? thank you.