Un-Mix: Rethinking Image Mixtures for Unsupervised Visual Representation Learning
Zhiqiang Shen1,3, Zechun Liu1, Zhuang Liu2, Marios Savvides1, Trevor Darrell2 and Eric Xing1,3
Abstract
The recently advanced unsupervised learning approaches use the siamese-like framework to compare two “views” from the same image for learning representations. Making the two views distinctive is a core to guarantee that unsupervised methods can learn meaningful information. However, such frameworks are sometimes fragile on overfitting if the augmentations used for generating two views are not strong enough, causing the over-confident issue on the training data. This drawback hinders the model from learning subtle variance and fine-grained information. To address this, in this work we aim to involve the soft distance concept on label space in the contrastive-based unsupervised learning task and let the model be aware of the soft degree of similarity between positive or negative pairs through mixing the input data space, to further work collaboratively for the input and loss spaces. Despite its conceptual simplicity, we show empirically that with the solution – Unsupervised image mixtures (Un-Mix), we can learn subtler, more robust and generalized representations from the transformed input and corresponding new label space. Extensive experiments are conducted on CIFAR-10, CIFAR-100,
中文速览
无监督对比学习（contrastive learning）通过让同一张图片的两个"视角"相互比较来学习图像表征，但当数据增强力度不足时，模型容易对训练数据过度自信，导致学到的特征缺乏细粒度信息。针对这一问题，作者提出了 Un-Mix 方法：在训练时将两张图片按比例混合（采用 Mixup 或 CutMix 策略）生成新的输入，同时相应地软化对比损失中的标签，让模型不再只区分"完全相同"或"完全不同"的样本对，而是感知它们之间连续的相似程度。在 CIFAR-10/100、STL-10、Tiny ImageNet 和 ImageNet 上，将 Un-Mix 应用于 SimCLR、BYOL、MoCo V1&V2、SwAV 等主流方法后，线性评估精度稳定提升 1–3%，下游目标检测任务上用 200 轮训练甚至超过了基线方法 800 轮的结果。这项工作说明同步操纵输入空间与标签空间是提升无监督表征质量的简单而有效的通用手段，只需数行代码即可嵌入现有框架。
原文 arXiv:2003.05438;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2003.05438v5