EffiScene: Efficient Per-Pixel Rigidity Inference for Unsupervised Joint Learning of Optical Flow, Depth, Camera Pose and Motion Segmentation.
Yang JiaoTrac D. TranGuangming ShiPublished in: CVPR (2021)
Keyphrases
- motion segmentation
- optical flow
- image sequences
- motion model
- motion estimation
- moving objects
- unsupervised learning
- point correspondences
- computer vision
- flow field
- motion field
- missing data
- d objects
- video sequences
- model selection
- viewpoint
- computational complexity
- machine learning
- dynamic scenes
- camera pose
- image motion
- feature tracking
- multibody