Can not find a use case.
Several samples to HF spaces and no enhancement was noticed.
Is this just doing a slight improvement to over linear/cubic upsampling?
Please describe your product's applicability. I find none in the time I spent so far.
@BingoBird Hi, could you provide a sample? It might not generalize perfectly since it was strictly trained on VCTK dataset to fairly compare against other baselines but I found it still should perform decently with out of domain samples. It should also be significantly better then linear/cubic/sinc upsampling, those almost provide 0 benefit in quality which can be measured with benchmarks.
Compare your approach to realesrgan please.
Compare your approach to realesrgan please.
satire ??
Most likely, I'm closing the issue. But feel free to upload any sample where the model performs poorly. Will try my best to fix it.