Closed Johnreidsilver closed 5 months ago
The "upscale image" node for onnx is really just a "run model" node. That being said though, not every model is guaranteed to work. We only officially support models converted from pytorch to onnx or ncnn on either of those frameworks.
As for what's supported by pytorch, we have to add support for models manually. You can't just find some random code on GitHub and expect the model to work. If you want it to be supported, create an issue in the spandrel repo: https://github.com/chaiNNer-org/spandrel
Hopefully that clears things up a bit
Thanks, been messing with different models and found one that works for dehazing: https://github.com/AmeryXiong/MixDehazeNet
edit: I see you already have in on the spandrel list
Hi, was trying out chaiNNer (thx btw! super simple to install in Ubuntu and excellent GUI), and it works fine with upscaling models, but what about other models like deblur, dehaze, and low light correction ? I can load them, but can't figure how to apply those.
For example: https://github.com/VITA-Group/EnlightenGAN
This one will fail on load with a "not supported" message. I tried with an ONNX version from here: https://github.com/PINTO0309/PINTO_model_zoo/tree/main/214_EnlightenGAN
It doesn't throw an error loading, but I can only connect it to either upscale
or remove bg that doesn't let the chain to be established with subtype "generic" not connectable to subtype "RemBg"
Do you plan to add support for this kind of "filtering" models or is it out of scope of chaiNNer ?