sczhou / upscale-a-video Goto Github PK

Could you please open source the degraded code of LQ video？

Hi @sczhou ,

Thanks for sharing the awesome work.

I have an awkward question: how did you create this kind of demo? Is there any out-of-the-box tool?

https://twitter.com/i/status/1734468279123775859

Peace, i hope you add option to choose good frames from the same video and from another high quality photos to be used as reference to enhance output quality and fidelity, this will be very important to enhance very bad quality videos, so many old videos have little number of high quality frames of faces and objects on videos, if we didn't use them as reference, the output video will look great but very different from original video, especially in faces, so if we use perfect frames from original video or from high quality photos from other sources, like high quality photos for the same person or even some photos from the same occasion like still photos from wedding party for example , this will increase the fidelity of the output video and reduce the AI hallucination very much, it's like references control nets, and ipadapters, and training loras and using roop, but we use them in every thing in the photos like faces, clothes, textures......etc, i know the program not released yet and it is may be in refining stage, but i hope you consider adding this feature, because it's results will be revolutionary🤞🤞🤞🤞

Peace, did Upscale-A-Video upscale one frame only or it use data from previous and next frames to increase quality and preserve coherence? and if it uses previous and next frames in upscaling, did it will be an option to choose how many frames will be used to enhance one frame? I'm looking forward to using the program soon.

Thank you for your open source YouHQ dataset!
But I'm confused about the dataset had been cropped to 80*80 for training mentioned in your paper, if you had used the full size YouHQ videos during training phase? And if only using 80*80 videos, how the model can adapt to larger resolution videos for super-resolution?

Looking forward to open source, I think if this work can achieve the effects mentioned in the README, it will be an important progress for the AI ​​video generation. @sczhou

Congrats on this great work!! When will the code be released? Thanks!