OD3: Optimization-free Dataset Distillation for Object Detection

2025-06-02

AI Summary

To address the lack of efficient, optimization-free methods for dataset distillation in object detection, this paper proposes OD3—the first optimization-free data distillation framework specifically designed for detection tasks. OD3 employs a two-stage strategy: (i) spatially aware, instance-level synthesis via iterative candidate instance localization, and (ii) confidence-driven instance selection using a frozen pre-trained observer model—both stages requiring no gradient-based optimization. The result is a compact, high-informativeness detection dataset. On MS COCO and PASCAL VOC, OD3 achieves competitive or superior detection performance using only 0.25%–5% of the original training data. Notably, at a 1.0% compression ratio on COCO, OD3 improves mAP₅₀ by over 14 percentage points compared to the full dataset baseline, significantly outperforming existing detection distillation and core-set methods and establishing new state-of-the-art performance in detection-specific data distillation.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Training large neural networks on large-scale datasets requires substantial computational resources, particularly for dense prediction tasks such as object detection. Although dataset distillation (DD) has been proposed to alleviate these demands by synthesizing compact datasets from larger ones, most existing work focuses solely on image classification, leaving the more complex detection setting largely unexplored. In this paper, we introduce OD3, a novel optimization-free data distillation framework specifically designed for object detection. Our approach involves two stages: first, a candidate selection process in which object instances are iteratively placed in synthesized images based on their suitable locations, and second, a candidate screening process using a pre-trained observer model to remove low-confidence objects. We perform our data synthesis framework on MS COCO and PASCAL VOC, two popular detection datasets, with compression ratios ranging from 0.25% to 5%. Compared to the prior solely existing dataset distillation method on detection and conventional core set selection methods, OD3 delivers superior accuracy, establishes new state-of-the-art results, surpassing prior best method by more than 14% on COCO mAP50 at a compression ratio of 1.0%. Code and condensed datasets are available at: https://github.com/VILA-Lab/OD3.

Problem

• Optimization-free dataset distillation for object detection
• Reducing computational demands in dense prediction tasks
• Synthesizing compact datasets for complex detection settings

Innovation

• Optimization-free dataset distillation for detection
• Two-stage candidate selection and screening
• Pre-trained observer model filters low-confidence objects