Interpreting Object-level Foundation Models via Visual Precision Search

2024-11-25

arXiv.org

AI Summary

Existing object-level foundation models (e.g., Grounding DINO, Florence-2) suffer from limited interpretability: gradient-based methods struggle with precise localization due to intricate multimodal fusion, while perturbation-based approaches yield noisy, coarse-grained saliency maps. This paper proposes Visual Precision Search (VPS), a gradient-free, black-box attribution method that requires no access to model parameters or internal multimodal architecture. VPS partitions the input into sparse subregions and leverages cross-modal consistency and collaboration scoring to accurately localize critical decision regions. Its key contributions include: (i) the first black-box attribution paradigm tailored for object-level models; (ii) theoretical error bounds on attribution fidelity; and (iii) support for failure-case diagnosis. Evaluated on RefCOCO, MS COCO, and LVIS, VPS outperforms state-of-the-art methods—improving grounding confidence by 20.1–31.6% for Grounding DINO and 66.9–102.9% for Florence-2. Code is publicly available.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Advances in multimodal pre-training have propelled object-level foundation models, such as Grounding DINO and Florence-2, in tasks like visual grounding and object detection. However, interpreting these models' decisions has grown increasingly challenging. Existing interpretable attribution methods for object-level task interpretation have notable limitations: (1) gradient-based methods lack precise localization due to visual-textual fusion in foundation models, and (2) perturbation-based methods produce noisy saliency maps, limiting fine-grained interpretability. To address these, we propose a Visual Precision Search method that generates accurate attribution maps with fewer regions. Our method bypasses internal model parameters to overcome attribution issues from multimodal fusion, dividing inputs into sparse sub-regions and using consistency and collaboration scores to accurately identify critical decision-making regions. We also conducted a theoretical analysis of the boundary guarantees and scope of applicability of our method. Experiments on RefCOCO, MS COCO, and LVIS show our approach enhances object-level task interpretability over SOTA for Grounding DINO and Florence-2 across various evaluation metrics, with faithfulness gains of 23.7%, 31.6%, and 20.1% on MS COCO, LVIS, and RefCOCO for Grounding DINO, and 102.9% and 66.9% on MS COCO and RefCOCO for Florence-2. Additionally, our method can interpret failures in visual grounding and object detection tasks, surpassing existing methods across multiple evaluation metrics. The code will be released at https://github.com/RuoyuChen10/VPS.

Problem

• Interpreting decisions of object-level foundation models
• Overcoming limitations in gradient and perturbation methods
• Enhancing interpretability for visual grounding and object detection

Innovation

• Visual Precision Search for accurate attribution maps
• Bypasses internal parameters to avoid fusion issues
• Uses consistency scores to identify critical regions