A Survey on Data Curation for Visual Contrastive Learning: Why Crafting Effective Positive and Negative Pairs Matters

2025-02-12

AI Summary

This work addresses the lack of systematic guidance in constructing positive and negative sample pairs for visual contrastive learning. We propose the first unified taxonomy for pair construction, rigorously delineating the applicability boundaries and coupling mechanisms among data augmentation, semantic alignment, and hard-negative mining paradigms. Our method integrates key techniques—including self-supervised augmentation, dynamic metric learning, memory-bank sampling, and cross-modal alignment—to establish a reproducible paired-evaluation benchmark and practical implementation guide. Extensive experiments demonstrate that the framework substantially improves representation quality, accelerates training convergence, enhances downstream transfer performance, and reduces computational overhead. By unifying theoretical analysis with engineering practice, our work provides both foundational principles and actionable methodologies for data governance in contrastive learning.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Visual contrastive learning aims to learn representations by contrasting similar (positive) and dissimilar (negative) pairs of data samples. The design of these pairs significantly impacts representation quality, training efficiency, and computational cost. A well-curated set of pairs leads to stronger representations and faster convergence. As contrastive pre-training sees wider adoption for solving downstream tasks, data curation becomes essential for optimizing its effectiveness. In this survey, we attempt to create a taxonomy of existing techniques for positive and negative pair curation in contrastive learning, and describe them in detail.

Problem

• Improving representation quality in visual contrastive learning
• Optimizing data pair curation for training efficiency
• Taxonomy of techniques for effective contrastive learning

Innovation

• Visual contrastive learning techniques
• Positive and negative pair curation
• Taxonomy of data curation methods