# RICAP - @takahashiDataAugmentationUsing2020 - Random image [Cropping](Cropping.md) and patching - [Cropping](Cropping.md) four regions from randomly sampled images and augmenting them to create a new image. - The generated image has mixed labels proportional to the pasted area - The area of cropped regions in the output image is determined by sampling through uniform distribution - anywhere-RICAP (origin can be anywhere), center-RICAP (origin can only be in the middle of the image), and corner-RICAP (origin can only be in corners - Corner-RICAP has shown the best performance because a larger region of one image is visible to the network to learn