F2Depth: Self-supervised indoor monocular depth estimation via optical flow consistency and feature map synthesis.
Xiaotong GuoHuijie ZhaoShuwei ShaoXudong LiBaochang ZhangPublished in: Eng. Appl. Artif. Intell. (2024)
Keyphrases
- depth estimation
- optical flow
- feature maps
- depth cues
- depth map
- image sequences
- stereo vision
- depth information
- stereo matching
- dynamic scenes
- ego motion
- depth estimates
- d scene
- self organizing maps
- real scenes
- scene understanding
- super resolution
- kernel function
- saliency map
- motion segmentation
- motion estimation
- motion field
- stereo pair
- high quality
- camera motion
- moving objects
- feature vectors
- motion model
- computer vision
- three dimensional
- input image
- real time
- video sequences
- high resolution
- multi view
- disparity map
- active learning
- stereo images
- feature space
- object boundaries
- scene flow
- image matching
- d objects
- support vector machine
- image data
- feature points
- specular reflection
- pose estimation
- depth from defocus