machine learning; image augmentation; augmentation consistency

Data shift is gap between data distribution used for training and data distribution encountered in real-world. Data augmentations help narrow the gap by generating new data samples, increasing data variability, and data space coverage. We present new data augmentation: Unproportinal mosaicing. Our augmentation randomly splits an image into various-sized blocks and swap their content (pixels) while maintaining block sizes. Our method achieves a lower error rate when combined with the other state-of-the-art augmentations.