TMT-VIS: Taxonomy-aware Multi-dataset Joint Training for Video Instance Segmentation.
Rongkun ZhengLu QiXi ChenYi WangKun WangYu QiaoHengshuang ZhaoPublished in: NeurIPS (2023)
Keyphrases
- video segmentation
- segmentation algorithm
- object segmentation and tracking
- segmentation method
- weakly labeled
- training dataset
- multiscale
- video sequences
- level set
- video frames
- video streams
- video data
- human actions
- image segmentation
- video content
- video dataset
- shape prior
- real time
- text detection
- temporal segmentation
- supervised learning
- joint estimation
- single frame
- segmentation accuracy
- foreground background segmentation
- neural network
- training process
- video clips
- multimedia
- training data
- video analysis
- region growing
- video scene
- computer vision
- layered representation
- image retrieval
- motion estimation
- training samples
- video surveillance
- video objects
- video search
- motion segmentation
- video retrieval
- object segmentation