Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error.
Stanislav FortAndrew BrockRazvan PascanuSoham DeSamuel L. SmithPublished in: CoRR (2021)
Keyphrases
- image data
- image features
- image content
- training set
- image segmentation
- image retrieval
- image representation
- image analysis
- image regions
- single image
- image classification
- input image
- error free
- image matching
- multiscale
- test images
- hough transform
- segmentation method
- error rate
- multiple objects
- segmentation algorithm
- training samples
- feature points
- low level
- data sets
- sample points
- motion estimation
- spatial information
- high resolution
- sample images
- image pixels
- template matching
- sample set
- reconstruction error
- horizontal and vertical directions
- image collections
- image set
- vector field
- test data
- training examples
- edge detection
- feature space
- similarity measure
- feature extraction