Zeus: Efficiently Localizing Actions in Videos using Reinforcement Learning.
Pramod ChunduriJaeho BangYao LuJoy ArulrajPublished in: CoRR (2021)
Keyphrases
- reinforcement learning
- perceptual aliasing
- human activities
- action selection
- video sequences
- action detection
- function approximation
- partially observable
- plan recognition
- state space
- multiagent reinforcement learning
- video frames
- human actions
- partially observable domains
- recognizing actions
- initial state
- model free
- optimal policy
- reward function
- video database
- dynamic scenes
- group activities
- temporal structure
- learned knowledge
- markov decision processes
- state and action spaces
- video data
- recognizing human actions
- mobile robot
- learning algorithm