Video BagNet: short temporal receptive fields increase robustness in long-term action recognition.
Ombretta StrafforelloXin LiuKlamer SchutteJan van GemertPublished in: CoRR (2023)
Keyphrases
- action recognition
- receptive fields
- human actions
- space time
- spatial temporal
- action classification
- motion history images
- spatial and temporal
- video dataset
- biologically inspired
- spatio temporal
- recognizing human actions
- action detection
- natural images
- human activities
- video sequences
- activity recognition
- temporal information
- space time interest points
- image representation
- recognition of human actions
- atomic actions
- computer vision
- temporal structure
- saliency map
- visual information
- input space
- video data
- video frames
- visual features
- multimedia
- high level
- temporal resolution
- multiscale
- neural network
- object recognition
- temporal relations
- pairwise
- low level
- higher order