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