SPARC: Score Prompting and Adaptive Fusion for Zero-Shot Multi-Label Recognition in Vision-Language Models

2025-02-24

AI Summary

Zero-shot multi-label recognition (MLR) faces challenges including the absence of training data and the inability to fine-tune vision-language models (VLMs). This paper proposes a fully black-box approach that requires no ground-truth labels, model fine-tuning, or architectural modifications—relying solely on raw VLM output scores. Our key contributions are: (1) an empirical discovery that the VLM’s second-highest class score exhibits higher discriminability than the top score for multi-label inference; (2) construction of composite prompts grounded in object co-occurrence priors; and (3) an unsupervised score debiasing and adaptive fusion mechanism that explicitly models semantic ambiguities inherent in “AND/OR” logical relations among labels. Evaluated across multiple benchmarks, our method consistently outperforms training-dependent zero-shot baselines, achieving substantial gains in mean average precision (mAP) while improving both multi-label ranking quality and prediction robustness.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Zero-shot multi-label recognition (MLR) with Vision-Language Models (VLMs) faces significant challenges without training data, model tuning, or architectural modifications. Existing approaches require prompt tuning or architectural adaptations, limiting zero-shot applicability. Our work proposes a novel solution treating VLMs as black boxes, leveraging scores without training data or ground truth. Using large language model insights on object co-occurrence, we introduce compound prompts grounded in realistic object combinations. Analysis of these prompt scores reveals VLM biases and ``AND''/``OR'' signal ambiguities, notably that maximum compound scores are surprisingly suboptimal compared to second-highest scores. We address these through a debiasing and score-fusion algorithm that corrects image bias and clarifies VLM response behaviors. Our method enhances other zero-shot approaches, consistently improving their results. Experiments show superior mean Average Precision (mAP) compared to methods requiring training data, achieved through refined object ranking for robust zero-shot MLR.

Problem

• Zero-shot multi-label recognition challenges
• Leveraging scores without training data
• Debiasing and score-fusion algorithm enhances accuracy

Innovation

• Zero-shot multi-label recognition
• Compound prompts without training
• Debiasing and score-fusion algorithm