zzmartin

I find it converge fast too, the loss looks great, but I can't use it to get a good result, wonder why?

Have u implemented it? I tried DDIM but found it fail in SR. I wonder how could it be implemented

> Were you able to get UCF101 unconditional training to work? I am getting moving colored patches for the sampled video yeah, just adjust unet and lr should work. I...

is your loss converged? and maybe train longer? I train with eyemakeup folder, and need 60-70k to get good 64*64 results, need 150+k to get good 128*128 results. And the...

lr = 1e-4, batch_size=8(2 1080ti), 10k cosine warmup, model = Unet3D(dim = 64,dim_mults = (1, 2, 4, 8, 8)) prob_focus_present I don't find matter a lot, num it 0.0 or...

well, it shouldn't be that slow. I can train 70k using less than 2 days.(with DP accelerate)

aha, i am sorry, i mean using dataparallel to accelerate. sorry make you confused

> Also your learning rate seems high compared to mine. Interesting 1e-04 for batch_size = 8 > > Original paper is 3e-04 for batch_size = 128 1e-4 is smaller than...

yeah, i'm just first 10k cosine warmup, then also linear down from 1e-4. I tried some bigger lr(5e-4) when producing 128\*128, finding that really hard to converge, and results are...

I use l1, not have time to try l2, I will give it a try!