FlowFormer++: Masked Cost Volume Autoencoding for Pretraining Optical Flow Estimation.
Xiaoyu ShiZhaoyang HuangDasong LiManyuan ZhangKa Chun CheungSimon SeeHongwei QinJifeng DaiHongsheng LiPublished in: CVPR (2023)
Keyphrases
- optical flow estimation
- optical flow
- optical flow computation
- coarse to fine
- occlusion detection
- flow field
- motion field
- multiscale
- motion model
- stereo matching
- variational methods
- motion discontinuities
- motion estimation
- estimation of optical flow
- multi frame
- flow estimation
- velocity field
- model selection
- image sequences