VITA: Video Instance Segmentation via Object Token Association.
Miran HeoSukjun HwangSeoung Wug OhJoon-Young LeeSeon Joo KimPublished in: NeurIPS (2022)
Keyphrases
- video segmentation
- object segmentation
- video objects
- multiple objects
- video scene
- video sequences
- bounding box
- image segmentation
- segmentation accuracy
- motion cues
- video frames
- video data
- level set
- multimedia
- medical images
- d objects
- segmentation algorithm
- segmentation method
- foreground background segmentation
- multiscale
- object tracking
- video analysis
- signed distance
- object shape
- moving objects
- fully unsupervised
- shape prior
- video content
- objects in video sequences
- energy function
- object model
- video clips
- video surveillance
- region growing
- pixel level
- object motion
- image regions
- final segmentation
- combining information from multiple
- video tracking
- object segmentation and tracking
- global shape
- object contours
- foreground and background
- intensity images
- superpixels
- target object
- spatial relations
- key frames
- spatial and temporal