方川

> Hi, it looks like an overfitting problem. How's the visualisation of the training split and the PSNR metrics on train/val splits? It makes sense that the more views are...

hi xiao, have you fixed this error? I met the same error msg when using accelerate integrated with deepspeed, and I cannot find useful solutions to fix this problem, do...

> @fangchuan I can definitely see some issues when doing cubemap -> equirectangular -> perspective view when dealing with low resolution cubemaps; loss of data being one. And your idea...

> Can you preprocess your cubemap offline and convert it to a high-resolution panorama (e.g., 1024x2048) for faster loading into your dataloader? Loading panorama and cropping arbitrary perspective views should...

hi all, I am using the zero-terminal SNR noise scheduler to fine tuning a multi-modal stable diffusion([Wonder3D](https://github.com/xxlong0/Wonder3D)), the loss is pretty large even if I can get reasonable generations in...