Training data-efficient image transformers & distillation through attention.
Hugo TouvronMatthieu CordMatthijs DouzeFrancisco MassaAlexandre SablayrollesHervé JégouPublished in: ICML (2021)
Keyphrases
- binary images
- input image
- training data
- image analysis
- image segmentation
- input data
- natural images
- template matching
- image content
- image data
- single image
- image pixels
- prior knowledge
- hough transform
- image collections
- image structure
- learning algorithm
- decision trees
- image retrieval
- keypoints
- data sets
- image features
- image classification
- image representation
- test images
- region of interest
- multiscale
- feature points
- low level
- active learning
- segmentation method
- edge detection
- image regions
- text classification
- object recognition
- similarity measure
- denoising