TMT-VIS: Taxonomy-aware Multi-dataset Joint Training for Video Instance Segmentation.
Rongkun ZhengLu QiXi ChenYi WangKun WangYu QiaoHengshuang ZhaoPublished in: CoRR (2023)
Keyphrases
- object detection
- object detectors
- object segmentation
- segmentation accuracy
- computer vision
- video segmentation
- segmentation algorithm
- level set
- video data
- image segmentation
- segmentation method
- video frames
- video content
- objects in video sequences
- text detection
- region growing
- multimedia
- multiscale
- video sequences
- test images
- real time
- weakly labeled
- training set
- joint estimation
- edge detection
- benchmark datasets
- video streams
- foreground background segmentation
- single frame
- video dataset
- pre trained
- temporal segmentation
- ground truth
- object segmentation and tracking
- video shots
- space time
- dynamic scenes
- video retrieval
- video surveillance
- human actions
- shape prior
- medical images
- human activities
- image regions