FYI: Flip Your Images for Dataset Distillation

TL;DR

This paper identifies bilateral equivalence in real image datasets, where discriminative parts distribute symmetrically across left and right halves, hindering dataset distillation from encoding fine-grained object details. It introduces FYI, a plug-and-play technique that augments synthetic data with horizontally flipped counterparts and optimizes them jointly to break left–right duplication without altering models or objectives. Empirically, FYI yields consistent improvements across CIFAR-10/100, Tiny-ImageNet, and ImageNet subsets when paired with state-of-the-art distillation methods, especially in 1 IPC settings and across architectures. By transferring richer semantics from real to synthetic images, FYI mitigates bilateral equivalence and enhances robustness and detail in synthesized samples.

Abstract

Dataset distillation synthesizes a small set of images from a large-scale real dataset such that synthetic and real images share similar behavioral properties (e.g, distributions of gradients or features) during a training process. Through extensive analyses on current methods and real datasets, together with empirical observations, we provide in this paper two important things to share for dataset distillation. First, object parts that appear on one side of a real image are highly likely to appear on the opposite side of another image within a dataset, which we call the bilateral equivalence. Second, the bilateral equivalence enforces synthetic images to duplicate discriminative parts of objects on both the left and right sides of the images, limiting the recognition of subtle differences between objects. To address this problem, we introduce a surprisingly simple yet effective technique for dataset distillation, dubbed FYI, that enables distilling rich semantics of real images into synthetic ones. To this end, FYI embeds a horizontal flipping technique into distillation processes, mitigating the influence of the bilateral equivalence, while capturing more details of objects. Experiments on CIFAR-10/100, Tiny-ImageNet, and ImageNet demonstrate that FYI can be seamlessly integrated into several state-of-the-art methods, without modifying training objectives and network architectures, and it improves the performance remarkably.

Figures (12)

• Figure 1: Comparisons of existing dataset distillation methods and our approach with the 1 IPC setting on CIFAR-100 cifar: Camel, chair, and lawn mower classes. (a) Objects in natural images are oriented diversely, and (b) current dataset distillation methods ((left) MTT mtt and (right) DSA dsa) synthesize symmetric images with repeated patterns in the left and right halves, neglecting fine-grained details of objects. (c) Applying FYI to MTT and DSA avoids this problem, while capturing the fine-grained details.
• Figure 2: Distributions of discriminative object parts for the class of (from left to right) tench, goldfish, white shark, tiger shark, and hammerhead, on ImageNet imagenet. We count how many times each pixel belongs to the top-10% of attention values obtained from class activation maps cam using a pre-trained ResNet-18 resnet. Red: high, Blue: low.
• Figure 3: The bilateral equivalence of a real-world dataset. We compute the unequalness score with a set of image(s) for each object class, where we randomly sample the images from CIFAR-10 cifar, using DC dc, DSA dsa and DM dm as distance metrics in \ref{['eq:4']}, and show the scores averaged over the classes.
• Figure 4: FYI augments synthetic images with the flipped counterparts to avoid the influence of the bilateral equivalence for dataset distillation.
• Figure 5: The bilateral equivalence of synthetic datasets with and without FYI on CIFAR-100 cifar. We compute the unequalness score of synthetic images during training for (a-b) DSA dsa and (c-d) DM dm. FYI achieves a higher unequalness score compared to the vanilla methods during training, implying that it enables encoding different semantics on different halves of images. More experiments for different datasets, methods, and compression ratios can be found in the supplementary material.