Data augmentation is a powerful technique to increase the diversity of data, which can effectively improve the generalization ability of neural networks in image recognition tasks. Recent mixing-based data augmentations have achieved great success by randomly cropping a patch from one image and pasting it on another. And some works explore to use of the saliency information of the image to guide the mixing. We systematically study the importance of the saliency information for mixing data, and find that the saliency information is not necessary for promoting the augmentation performance. Furthermore, the mixing-based data mixing methods carry two problems of object information missing and label misallocation. We propose an effective and very easily implemented method, namely ResizeMix, which can mix data with preserved object information and true labels. We mix the data by directly resizing the source image to a small patch and paste it on another image. The obtained patch preserves more substantial object information compared with conventional cutting-based methods. ResizeMix achieves superior performance on both image classification and object detection tasks without additional computation cost.