GridMask Data Augmentation
This addresses data augmentation for computer vision practitioners, offering a simpler and more effective alternative to computationally expensive methods like AutoAugment.
The paper tackles the problem of data augmentation in computer vision by proposing GridMask, a novel information removal method that achieves state-of-the-art results. It outperforms AutoAugment on ImageNet, COCO2017, and Cityscapes datasets with notable performance improvements.
We propose a novel data augmentation method `GridMask' in this paper. It utilizes information removal to achieve state-of-the-art results in a variety of computer vision tasks. We analyze the requirement of information dropping. Then we show limitation of existing information dropping algorithms and propose our structured method, which is simple and yet very effective. It is based on the deletion of regions of the input image. Our extensive experiments show that our method outperforms the latest AutoAugment, which is way more computationally expensive due to the use of reinforcement learning to find the best policies. On the ImageNet dataset for recognition, COCO2017 object detection, and on Cityscapes dataset for semantic segmentation, our method all notably improves performance over baselines. The extensive experiments manifest the effectiveness and generality of the new method.