Disentangling Controllable Object through Video Prediction Improves Visual Reinforcement Learning.
Yuanyi ZhongAlexander G. SchwingJian PengPublished in: CoRR (2020)
Keyphrases
- reinforcement learning
- visual objects
- video data
- visual input
- visual analysis
- spatial relations
- prediction accuracy
- visual cues
- object motion
- video streams
- visual data
- video sequences
- machine learning
- object model
- d objects
- prediction model
- moving objects
- object detection and tracking
- visual information
- video analysis
- real time
- visual appearance
- space time
- learning algorithm
- reinforcement learning algorithms
- low level
- category specific
- video frames
- state space
- object tracking
- contextual cues
- visual concepts
- video clips
- multimedia
- video retrieval
- video surveillance
- video database
- visual features
- news video
- combining information from multiple
- stationary camera
- multi agent
- dynamic programming
- markov decision processes
- scalable video coding
- video content
- video objects
- function approximation
- multiple objects
- bounding box