Model seems to produce very over-sharpened results

hzwer / WACV2024-SAFA

WACV2024 - Scale-Adaptive Feature Aggregation for Efficient Space-Time Video Super-Resolution

MIT License

125 stars 12 forks source link

Model seems to produce very over-sharpened results #2

Open KLC04 opened 10 months ago

KLC04 commented 10 months ago

The provided pre- train seems to have an issue of significantly over sharpening while not adding any usable detail. Is this a known issue in the works?

hzwer commented 10 months ago

We recently tried to study models for visual effects. Current experimental models do not perform well on videos that appear to be synthesized differently from the training data. We are improving data augmentation refer to real-ESRGAN.

hzwer commented 10 months ago

https://github.com/hzwer/Practical-RIFE/tree/main#video-enhancement

KLC04 commented 10 months ago

Thanks for updating me! In my understanding this is for enhancement right, so no uspcaling and no interpolation?

hzwer commented 10 months ago

Yes, I hope to eliminate the influence of other factors and study the real degradation first.