In the training phase of CNN, we randomly choose a rectangle region in the image, and erase its pixels with random values.

1 Introduction Figure 1: Examples of Random Erasing. Random Erasing is parameter learning free, easy to implement, and can be integrated into most of the CNN-based recognition models.Īlbeit simple, Random Erasing yields consistent improvement in image classification, object detection and person re-identification (re-ID).įor image classification, our method improves WRN-28-10: top-1 error rate from 3.72% to 3.08% on CIFAR10, and from 18.68% to 17.65% on CIFAR100.įor object detection on PASCAL VOC 2007, Random Erasing improves Fast-RCNN from 74.8% to 76.2% in mAP.įor person re-ID, when using Random Erasing in recent deep models, we achieve the state-of-the-art accuracy: the rank-1 accuracy is 89.13% for Market-1501, 84.02% for DukeMTMC-reID, and 63.93% for CUHK03 under the new evaluation protocol. In this process, training images with various levels of occlusion are generated, which reduce the risk of network overfitting and make the model robust to occlusion. In training phase, Random Erasing randomly selects a rectangle region in an image, and erases its pixels with random values. In this paper, we introduce Random Erasing, a simple yet effective data augmentation techniques for training the convolutional neural network (CNN).

Free pascal random 255