Zhihang Zhong comments

Results 18 comments of


                                            Zhihang Zhong

Request for a practical Pre-BiT++ model trained on perceptual loss

Thank you for your enthusiasm! I am currently working on ways to improve the theoretical upper bound and practical performance of the video frame interpolation algorithms, which include RIFE and...

求一个国内的dataset链接，如果方便的话

不好意思，暂时只有google drive版本。你也可以选择用提供的code自己生成数据库。

谢谢答复～ > 它是基于 RIFEm 的面向应用版本 RIFEm是arbitrary time的吧？我看RIFE_HDv3是fixed time版本，为什么说是基于RIFEm的呢？我的观察是arbitrary time模型的效果普遍不如fixed time模型通过iteration插值得到结果的效果。请问arbitrary time除了可以指定时间以外，还有什么优势呢？ > pyramid 多一个阶段我发现最新版本，无论是refinenet，还是contextnet，unet都去掉了，只剩下3个IFBlock。意思是之前的contextnet和unet对实际视觉效果也没啥帮助吗？

Channel Similarity

Please refer to some publicly available repos for CKA calculation like this: https://github.com/yuanli2333/CKA-Centered-Kernel-Alignment

Request for [D,R] RIFE trained on Style loss, also called Gram matrix loss (the best perceptual loss function)

Hi, thanks for the advice and sorry for the late reply. I can try it when I'm not busy. But if you or anyone can help to run this experiment,...

A minor issue with the config .yaml files

Hi, replacing the path with yours should work.

Testing with a custom video

Regarding "animation-from-blur", please refer to "valid_video.py". **Considering that you have a video, I would suggest using BiT, which is better suited and more convenient. You can find BiT at the...

Zhihang Zhong