Few-Shot Synthetic Data Generation with Diffusion Models for Downstream Vision Tasks

TL;DR

This paper introduces a lightweight synthetic data augmentation method using diffusion models fine-tuned with LoRA adapters, significantly improving rare class recognition in visual tasks with minimal real data.

Contribution

It demonstrates a scalable approach for rare class augmentation by fine-tuning diffusion models with few real images, enhancing performance across diverse visual domains.

Findings

Synthetic augmentation improves rare-class recall and F1 scores.

Moderate synthetic data yields optimal performance, with diminishing returns at higher ratios.

The approach is effective in both medical and industrial visual recognition tasks.

Abstract

Class imbalance is a persistent challenge in visual recognition, particularly in safety-critical domains where collecting positive examples is expensive and rare events are inherently underrepresented. We propose a lightweight synthetic data augmentation pipeline that fine-tunes a LoRA adapter on as few as 20-50 real images of a rare class and uses a pretrained diffusion model to generate synthetic samples for training. We systematically vary the synthetic-to-real ratio and evaluate the approach across two structurally different domains: chest X-ray pathology classification (NIH ChestX-ray14) and industrial surface crack detection (Magnetic Tile Defect dataset). All evaluations are performed on held-out sets of real images only. Across both domains, synthetic augmentation consistently improves rare-class recall and F1 compared to training with real data alone. Performance improves with moderate synthetic augmentation and shows diminishing returns as the synthetic ratio increases. These results suggest that LoRA-adapted diffusion models provide a simple and scalable mechanism for augmenting rare classes, enabling effective learning in data-scarce scenarios across heterogeneous visual domains.