RandomCrop returns uneven cropping for small images (e.g. MNIST and CIFAR-10/100), thus yielding sub-optimal performance in training.
For example, cropping 32x32 images from 36x36 images should have bound offsets [0, 1, 2, 3, 4]. However, currently only [0, 1, 2, 3] is used.
Here are 20 images generated using RandomCrop with DataAugmentation v0.2.8. The original images (32x32 in size) are padded by 2 pixels on each edge, randomly cropped to 32x32, and then up-sampled to 128x128 for display.
The left and top edges of the images are randomly padded by either 0/1/2 pixels, while the right edges are only paded by 0/1 pixels.
RandomCrop returns uneven cropping for small images (e.g. MNIST and CIFAR-10/100), thus yielding sub-optimal performance in training.
For example, cropping 32x32 images from 36x36 images should have bound offsets [0, 1, 2, 3, 4]. However, currently only [0, 1, 2, 3] is used.
Here are 20 images generated using RandomCrop with DataAugmentation v0.2.8. The original images (32x32 in size) are padded by 2 pixels on each edge, randomly cropped to 32x32, and then up-sampled to 128x128 for display.
The left and top edges of the images are randomly padded by either 0/1/2 pixels, while the right edges are only paded by 0/1 pixels.