Fake it till you make it: Learning transferable representations from synthetic ImageNet clones

TL;DR

This paper investigates whether synthetic images generated by models like Stable Diffusion can replace real images for training image classification models, showing promising results in closing performance gaps and enabling effective transfer learning.

Contribution

It demonstrates that synthetic ImageNet clones can significantly reduce the need for real images in training classifiers and exhibit strong transfer learning capabilities.

Findings

Synthetic ImageNet clones close the performance gap with real images.

Models trained on synthetic data generalize well to real-world tasks.

Minimal prompt engineering suffices to generate useful synthetic datasets.

Abstract

Recent image generation models such as Stable Diffusion have exhibited an impressive ability to generate fairly realistic images starting from a simple text prompt. Could such models render real images obsolete for training image prediction models? In this paper, we answer part of this provocative question by investigating the need for real images when training models for ImageNet classification. Provided only with the class names that have been used to build the dataset, we explore the ability of Stable Diffusion to generate synthetic clones of ImageNet and measure how useful these are for training classification models from scratch. We show that with minimal and class-agnostic prompt engineering, ImageNet clones are able to close a large part of the gap between models produced by synthetic images and models trained with real images, for the several standard classification benchmarks that we consider in this study. More importantly, we show that models trained on synthetic images exhibit strong generalization properties and perform on par with models trained on real data for transfer. Project page: https://europe.naverlabs.com/imagenet-sd/