Video deblurring

[Terka]

Hi, anybody tried some of new approaches like:
https://github.com/sczhou/STFAN
Would it be possible to implement to avisynth?

[kedautinh12]

If this plugin use PyTorch, i think can port to vapoursynth only

[poisondeathray]

I tested a few of them on other material for inferencing (not in the training or testing datasets to see how the pretrained models work in "real" scenarios). Some of the video ones are interesting , but take lots of GPU memory. The quality is generally worse if tiles are used for deblurring tasks (split video into temporal:height:width "tiles" to reduce memory consumption , with overlapping segments to reduce tile artifact errors, stitch back together)

The majority are single image (as opposed to video or multiple image) and deal with motion deblurring (defocus deblurring is the other category)

Many are difficult to run, or I haven't got them to work under windows. The majority use python - I doubt any would be possible in avisynth directly (maybe someday a python interpreter plugin would be possible...one can dream :D)

A partial list here
https://paperswithcode.com/task/image-deblurring

Some rankings by datasets (click on the datasets to see the graphs)
https://paperswithcode.com/task/deblurring

[kedautinh12]

Lol, "one can dream" :D