Mixup is a powerful data augmentation method that interpolates between two or
more examples in the input or feature space and between the corresponding
target labels. Many recent mixup methods focus on cutting and pasting two or
more objects into one image, which is more about efficient processing than
interpolation. However, how to best interpolate images is not well defined. In
this sense, mixup has been connected to autoencoders, because often
autoencoders "interpolate well", for instance generating an image that
continuously deforms into another.
In this work, we revisit mixup from the interpolation perspective and
introduce AlignMix, where we geometrically align two images in the feature
space. The correspondences allow us to interpolate between two sets of
features, while keeping the locations of one set. Interestingly, this gives
rise to a situation where mixup retains mostly the geometry or pose of one
image and the texture of the other, connecting it to style transfer. More than
that, we show that an autoencoder can still improve representation learning
under mixup, without the classifier ever seeing decoded images. AlignMix
outperforms state-of-the-art mixup methods on five different benchmarks.

介绍了一种名为 AlignMix 的新型混合方法，通过在特征空间中几何对齐两个图像，从而使混合方法保留一个图像的几何特征和另一个图像的纹理，提高了表示学习效果。