DataDream: Few-shot Guided Dataset Generation

TL;DR

DataDream is a novel framework that synthesizes high-quality, in-distribution training datasets from few-shot examples, significantly improving downstream image classification accuracy across multiple datasets.

Contribution

It introduces a method to fine-tune diffusion models with LoRA on few-shot data, enabling more faithful synthetic dataset generation for classification tasks.

Findings

Outperforms previous methods on 7 out of 10 datasets in classification accuracy

Generates more in-distribution and fine-grained images for training

Provides insights into factors affecting synthetic data quality and model performance

Abstract

While text-to-image diffusion models have been shown to achieve state-of-the-art results in image synthesis, they have yet to prove their effectiveness in downstream applications. Previous work has proposed to generate data for image classifier training given limited real data access. However, these methods struggle to generate in-distribution images or depict fine-grained features, thereby hindering the generalization of classification models trained on synthetic datasets. We propose DataDream, a framework for synthesizing classification datasets that more faithfully represents the real data distribution when guided by few-shot examples of the target classes. DataDream fine-tunes LoRA weights for the image generation model on the few real images before generating the training data using the adapted model. We then fine-tune LoRA weights for CLIP using the synthetic data to improve downstream image classification over previous approaches on a large variety of datasets. We demonstrate the efficacy of DataDream through extensive experiments, surpassing state-of-the-art classification accuracy with few-shot data across 7 out of 10 datasets, while being competitive on the other 3. Additionally, we provide insights into the impact of various factors, such as the number of real-shot and generated images as well as the fine-tuning compute on model performance. The code is available at https://github.com/ExplainableML/DataDream.