Unsupervised Learning of Depth, Optical Flow and Pose With Occlusion From 3D Geometry.
Guangming WangChi ZhangHesheng WangJingchuan WangYong WangXinlei WangPublished in: IEEE Trans. Intell. Transp. Syst. (2022)
Keyphrases
- unsupervised learning
- optical flow
- occlusion boundaries
- depth ordering
- human pose
- pose variations
- partial occlusion
- pose estimation
- depth images
- supervised learning
- image sequences
- motion discontinuities
- motion segmentation
- ego motion
- illumination variations
- flow field
- scene geometry
- optical flow fields
- d scene
- cluttered background
- depth cues
- depth map
- object boundaries
- planar objects
- image motion
- semi supervised
- optical flow estimation
- motion field
- optical flow algorithm
- depth information
- three dimensional
- d objects
- object shape
- motion estimation
- motion parameters
- machine learning
- object recognition
- reinforcement learning
- computer vision
- velocity field
- feature tracking
- model selection
- camera motion
- action recognition
- human body
- expectation maximization
- moving objects
- optical flow computation
- probabilistic model
- dimensionality reduction
- motion cues
- motion model