hzwer
I don't quite understand, what should I write?
1. You can first convert it to png, and then compose it into a video. I don't have much experience with codecs, perhaps you would be better off using existing...
因为主要是在计算一个可以放缩尺度的光流场，所以不一定要全图输入来计算 类似于图片超分模型
https://www.crcv.ucf.edu/data1/UCF101-DS/
I am experimenting with an idea similar to https://github.com/zzh-tech/InterpAny-Clearer, and it is not debugged yet. I hope that I will provide the code in about one or two weeks.
没有啊 我代码完全直接打包发出来的
batchsize 8 会不会少训练了很多
对 但是加完后效果反而下降了 有点怪
应该是同个验证集；单机多卡的
这个测试是 vimeo 吧，可能只有加同源会看到指标提升 真正实质指标还是得找一些综合的OOD数据