Does CLIP Bind Concepts? Probing Compositionality in Large Image Models.
Martha LewisQinan YuJack MerulloEllie PavlickPublished in: CoRR (2022)
Keyphrases
- image data
- single image
- input image
- multiscale
- probabilistic model
- visual concepts
- image representation
- low level
- image analysis
- image features
- image regions
- image segmentation
- image content
- edge detection
- bayesian framework
- image classification
- random fields
- hough transform
- image collections
- lighting conditions
- template matching
- region of interest
- vector field
- object detection
- image matching
- prior knowledge
- image pixels
- similarity measure
- high resolution
- image retrieval
- image structure
- statistical model
- keypoints
- segmentation method
- high level
- object recognition