qiuyu96 / CoDeF

[CVPR'24 Highlight] Official PyTorch implementation of CoDeF: Content Deformation Fields for Temporally Consistent Video Processing
https://qiuyu96.github.io/CoDeF/
Other
4.84k stars 385 forks source link

重建视频很清楚,但是canonical图片很模糊 #70

Open lhb365 opened 1 year ago

lhb365 commented 1 year ago

你好,我按照例程进行了模型的训练,并提高了提高annealed步数和训练步数,训练之后的模型可以很好的重建原视频,但是canonical图片很模糊,请问是哪方面的原因呢?

Me-Maped commented 1 year ago

俺也一样,视频内运动幅度比较大的情况会这样