PatchBlender: A Motion Prior for Video Transformers.
Gabriele PratoYale SongJanarthanan RajendranR. Devon HjelmNeel JoshiSarath ChandarPublished in: CoRR (2022)
Keyphrases
- space time
- video sequences
- object motion
- temporal filtering
- spatial and temporal
- moving camera
- key frames
- motion features
- input video
- video data
- motion analysis
- static images
- visual cues
- low frame rate
- motion estimation
- successive frames
- video frames
- temporal coherence
- multimedia
- image sequences
- moving objects
- shot change detection
- dynamic scenes
- video representation
- surveillance videos
- video footage
- temporal continuity
- video analysis
- motion detection
- dynamic textures
- real time
- video clips
- camera motion
- video content
- motion model
- motion parameters
- video streams
- object detection and tracking
- video retrieval
- temporal structure
- video signals
- video summarization
- motion trajectories
- temporal domain
- motion capture data
- visual data
- prior knowledge
- video processing
- motion patterns
- visual information
- human motion
- multiview video
- stationary camera
- feature points
- motion field
- human actions
- motion segmentation
- temporal consistency
- video database
- spatio temporal
- layered representation
- optical flow
- wide area motion imagery
- video coding
- camera movement
- super resolution
- visual features
- video scene
- video shots
- image frames