You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
These models are very slow (significantly slower than waifu2x.swin_unet_*), even with TRT backend.
These models are also poorly supported by various backends. With dynamic shapes, TRT usually reports internal error. Naive fp16 quantization affects accuracy.
As a result, these models are not included in built-in models. What's more, other transformer models may have already surpassed them in terms of reconstruction quality (e.g. vs-grlir, SRFormer).
I am using Hybrid developer version with vsmlrt addon. Could you please share vs-grlir or SRFormer onnx models, if vsmlrt supports them? thanks you very much.
https://github.com/XPixelGroup/HAT
https://github.com/JingyunLiang/SwinIR
please, please?
The text was updated successfully, but these errors were encountered: