Misalign, Contrast then Distill: Rethinking Misalignments in Language-Image Pretraining.
Bumsoo KimYeonsik JoJinhyung KimSeung Hwan KimPublished in: CoRR (2023)
Keyphrases
- image data
- input image
- single image
- image features
- contrast enhancement
- image representation
- template matching
- multiscale
- image content
- image classification
- image segmentation
- image pixels
- high resolution
- image analysis
- image retrieval
- image matching
- image collections
- segmentation algorithm
- test images
- segmentation method
- region of interest
- pixel values
- grey level
- language learning
- spatial information
- programming language
- object recognition
- similarity measure