Reinforcement Learning for Weakly Supervised Temporal Grounding of Natural Language in Untrimmed Videos.
Jie WuGuanbin LiXiaoguang HanLiang LinPublished in: CoRR (2020)
Keyphrases
- weakly supervised
- reinforcement learning
- natural language
- weakly labeled
- relation extraction
- superpixels
- topic models
- object class
- information extraction
- machine learning
- semi supervised
- video frames
- video sequences
- co occurrence
- named entities
- question answering
- natural language processing
- domain specific
- state space
- knowledge representation
- viewpoint
- d objects
- bounding box
- automatic extraction
- training data
- image sequences
- computer vision