shallweiwei

I have noticed the output is sometimes really bad on 256 resolution

2

When I change the resolution to 320 or more,it's much better.Does anyone have the same problem?

Loading model is very slow for the first time!

1

I tried Controlnet for two days. I removed the visual interface.But I found that the model needs to be loaded for a long time when I run it for the...

how to use the pretrained model?

3

I have added resume_from_checkpoint in the yaml,but the output of the first epoch is still 0

abrupt masks

4

I use my dataset to train and notice that the masks usually have different colors than the parts around them.They are easy to distinguish with naked eyes. So if there...

It seems SwinIR doesn't use patch merging.

3

Excellent work! It seems that there is not patch merging.So it means different windows can't exchange information.I understand you don't want to downsample,but maybe the windows should connect with each...

fix the ConvTranspose2d bug,whose dilation parameter should be the last one

I noticed that the instance segmentation metric is very high, and I also want to try training, but it seems that the annotation below says "(test)", which means it can...

Hello, the synthetic data looks great, can you open source the blender code?