Scaling Laws of Synthetic Images for Model Training ... for Now

TL;DR

This paper investigates how synthetic images generated by state-of-the-art text-to-image models scale for training vision systems, revealing their limitations and potential benefits in specific scenarios such as limited data or out-of-distribution tasks.

Contribution

It provides a detailed analysis of the scaling laws of synthetic images for training classifiers and CLIP models, highlighting factors affecting performance and identifying scenarios where synthetic data is most beneficial.

Findings

Synthetic images follow a scaling trend similar to real images in CLIP training.

Synthetic images underperform in scaling for supervised image classifiers.

Limitations of current text-to-image models hinder their effectiveness for classifier training.

Abstract

Recent significant advances in text-to-image models unlock the possibility of training vision systems using synthetic images, potentially overcoming the difficulty of collecting curated data at scale. It is unclear, however, how these models behave at scale, as more synthetic data is added to the training set. In this paper we study the scaling laws of synthetic images generated by state of the art text-to-image models, for the training of supervised models: image classifiers with label supervision, and CLIP with language supervision. We identify several factors, including text prompts, classifier-free guidance scale, and types of text-to-image models, that significantly affect scaling behavior. After tuning these factors, we observe that synthetic images demonstrate a scaling trend similar to, but slightly less effective than, real images in CLIP training, while they significantly underperform in scaling when training supervised image classifiers. Our analysis indicates that the main reason for this underperformance is the inability of off-the-shelf text-to-image models to generate certain concepts, a limitation that significantly impairs the training of image classifiers. Our findings also suggest that scaling synthetic data can be particularly effective in scenarios such as: (1) when there is a limited supply of real images for a supervised problem (e.g., fewer than 0.5 million images in ImageNet), (2) when the evaluation dataset diverges significantly from the training data, indicating the out-of-distribution scenario, or (3) when synthetic data is used in conjunction with real images, as demonstrated in the training of CLIP models.