M^2ConceptBase: A Fine-Grained Aligned Concept-Centric Multimodal Knowledge Base

2023-12-16

AI Summary

Existing knowledge bases exhibit coarse-grained alignment between images and textual descriptions, hindering multimodal models’ understanding of fine-grained visual concepts. Method: We introduce VLConceptKB—the first vision-language knowledge base centered on conceptual semantics—comprising 152K fine-grained concepts and 951K precisely aligned image–text triples, enabling concept-level cross-modal semantic grounding. We propose a “concept-centric” modeling paradigm and design a context-aware multimodal symbolic grounding method, achieving >95% alignment accuracy among concepts, images, and descriptions; we further establish a joint construction and human verification framework. Contribution/Results: Experiments demonstrate that VLConceptKB significantly improves VQA model performance on OK-VQA and effectively enhances multimodal large models’ capabilities in fine-grained concept understanding via retrieval augmentation.

Abstract

Multimodal knowledge bases (MMKBs) provide cross-modal aligned knowledge crucial for multimodal tasks. However, the images in existing MMKBs are generally collected for entities in encyclopedia knowledge graphs. Therefore, detailed groundings of visual semantics with linguistic concepts are lacking, which are essential for the visual concept cognition ability of multimodal models. Addressing this gap, we introduce M^2ConceptBase, the first concept-centric MMKB. M^2ConceptBase models concepts as nodes with associated images and detailed textual descriptions. We propose a context-aware multimodal symbol grounding approach to align concept-image and concept-description pairs using context information from image-text datasets. Comprising 951K images and 152K concepts, M^2ConceptBase links each concept to an average of 6.27 images and a single description, ensuring comprehensive visual and textual semantics. Human studies confirm more than 95% alignment accuracy, underscoring its quality. Additionally, our experiments demonstrate that M^2ConceptBase significantly enhances VQA model performance on the OK-VQA task. M^2ConceptBase also substantially improves the fine-grained concept understanding capabilities of multimodal large language models through retrieval augmentation in two concept-related tasks, highlighting its value.

Problem

• Knowledge Base
• Image-Text Alignment
• Multimodal Tasks

Innovation

• M^2ConceptBase
• Concept-centric Knowledge Base
• Multimodal Integration