Dyna-DepthFormer: Multi-frame Transformer for Self-Supervised Depth Estimation in Dynamic Scenes.
Songchun ZhangChunhui ZhaoPublished in: CoRR (2023)
Keyphrases
- multi frame
- depth estimation
- dynamic scenes
- motion estimation
- motion segmentation
- image sequences
- point correspondences
- video sequences
- super resolution
- multi view
- multiple views
- space time
- moving objects
- optical flow
- background subtraction
- motion analysis
- optic flow
- video frames
- moving camera
- stereo correspondence
- frame rate
- spatio temporal
- motion model
- three dimensional
- high resolution
- motion compensation
- state space
- motion vectors
- reinforcement learning