IFRNET models are not supported

chaiNNer-org / spandrel

Spandrel gives your project support for various PyTorch architectures meant for AI Super-Resolution, restoration, and inpainting. Based on the model support implemented in chaiNNer.

MIT License

139 stars 12 forks source link

IFRNET models are not supported #235

Open hooke007 opened 1 year ago

hooke007 commented 1 year ago

Information:

Chainner version: Alpha v0.18.7
OS: Windows 10

Description As title's description. Models are from here

Logs logs.zip

joeyballentine commented 1 year ago

This is not a bug. Chainner just doesn't support this model architecture yet.

theflyingzamboni commented 1 year ago

@joeyballentine IFRNet is a frame interpolation model. Is that something we ever plan to support in chaiNNer?

joeyballentine commented 1 year ago

Oh, i didn't realize that.

No, we probably won't support that for a really really long time. I do want to, but we'd need to change a bunch of stuff first

ferrosa commented 1 year ago

Even though this has been closed for a while, I'd like to comment so the relevant devs are aware.

IFRNet not just makes videos look better through higher fps, but also greatly improves the quality of subsequent ffmpeg temporal noise filters. It's a massive reduction of "darkness flickering" on phone recordings in low light conditions, and something I wouldn't want to do without anymore.

Looking to migrate my script-based workflow to chaiNNer, but this is a blocker.

joeyballentine commented 1 year ago

I can reopen this, but we aren't going to be supporting frame interpolation models until we rework how iteration works, otherwise it'll be a giant pain