Label-Consistent Dataset Distillation with Detector-Guided Refinement

2025-07-17

AI Summary

Dataset distillation (DD) often suffers from label inconsistency and loss of image details in synthetic samples, leading to degraded downstream performance. To address this, we propose a detector-guided diffusion distillation framework: first, a pretrained object detector identifies and rectifies low-confidence or mislabeled synthetic samples; second, a classification-detection dual-prototype-driven diffusion model generates high-fidelity candidate images; finally, a confidence-weighted and diversity-aware selection mechanism constructs a compact, robust surrogate dataset. Our method significantly improves image fidelity and label accuracy under extremely small-scale settings (e.g., 10–50 samples per class), achieving state-of-the-art downstream generalization on benchmarks including CIFAR and ImageNet. Moreover, it reduces both storage overhead and training cost compared to prior approaches.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Dataset distillation (DD) aims to generate a compact yet informative dataset that achieves performance comparable to the original dataset, thereby reducing demands on storage and computational resources. Although diffusion models have made significant progress in dataset distillation, the generated surrogate datasets often contain samples with label inconsistencies or insufficient structural detail, leading to suboptimal downstream performance. To address these issues, we propose a detector-guided dataset distillation framework that explicitly leverages a pre-trained detector to identify and refine anomalous synthetic samples, thereby ensuring label consistency and improving image quality. Specifically, a detector model trained on the original dataset is employed to identify anomalous images exhibiting label mismatches or low classification confidence. For each defective image, multiple candidates are generated using a pre-trained diffusion model conditioned on the corresponding image prototype and label. The optimal candidate is then selected by jointly considering the detector's confidence score and dissimilarity to existing qualified synthetic samples, thereby ensuring both label accuracy and intra-class diversity. Experimental results demonstrate that our method can synthesize high-quality representative images with richer details, achieving state-of-the-art performance on the validation set.

Problem

• Improve label consistency in distilled datasets
• Enhance structural detail in synthetic samples
• Optimize dataset distillation with detector guidance

Innovation

• Detector-guided refinement for label consistency
• Diffusion model generates multiple candidate images
• Joint selection based on confidence and diversity