Transformers Provably Learn Feature-Position Correlations in Masked Image Modeling.
Yu HuangZixin WenYuejie ChiYingbin LiangPublished in: CoRR (2024)
Keyphrases
- image features
- template matching
- single image
- image analysis
- input image
- image content
- image data
- image classification
- hough transform
- multiscale
- lighting conditions
- position and orientation
- image segmentation
- image retrieval
- test images
- segmentation method
- image collections
- relative position
- image representation
- feature points
- pixel values
- feature representation
- edge detection
- invariant features
- feature detectors
- high resolution
- spatial correlation