Bongard-OpenWorld: Few-Shot Reasoning for Free-form Visual Concepts in the Real World.
Rujie WuXiaojian MaZhenliang ZhangWei WangQing LiSong-Chun ZhuYizhou WangPublished in: ICLR (2024)
Keyphrases
- free form
- visual concepts
- video content
- data sets
- semantic concepts
- visual features
- video sequences
- object categories
- image collections
- knowledge base
- image annotation
- image content
- visual content
- complex scenes
- video shots
- positive examples
- video data
- feature vectors
- key frames
- training data
- semantic gap
- three dimensional
- reinforcement learning