Yzmblog / MonoHuman

MonoHuman: Animatable Human Neural Field from Monocular Video (CVPR 2023)
133 stars 9 forks source link

关于Forward Correspondence Search Module的理解 #3

Closed Time-Lord12th closed 1 year ago

Time-Lord12th commented 1 year ago

你好,关于Forward Correspondence Search Module,我想请问一下,训练过程中,是对于单目视频的每一帧,都会去寻找一张keyframe,然后拿这两帧的image feature指导nerf的渲染吗?

Yzmblog commented 1 year ago

你好,是的,不过我们的关键帧是离线选择好的。