Why is the training converges so slow?

[jin-cao-tma]

I have trained SST for several hours on Chikusei Dataset with RTX 4090Ti, but the PSNR is still very low(about 13) and the MSELoss decrease slowly. But within the same setting, T3SC can converge in about an hour. Is this normal?

[MyuLi]

It’s quite strange. Maybe you should decrease the learning rate. Or use some different initialization strategies