Name of the Dataset: M2ConceptBase: the first concept-centric multimodal concept knowledge base
Dataset Introduction: Traditional multimodal knowledge bases are typically entity-centric. To address the limitations of existing multimodal knowledge bases in aligning visual semantics with language concepts and to provide multimodal concept knowledge resources for the fine-grained long-tail concept understanding of multimodal large models, we introduce the first concept-centric multimodal concept knowledge base, M2ConceptBase. By employing a concept modeling strategy and achieving high-precision alignment with relevant images and textual descriptions, M2ConceptBase offers fine-grained cross-modal knowledge. Specifically, M2ConceptBase encompasses 152K concepts and 951K images, with an average of 6.27 images and corresponding descriptions associated with each concept, ensuring comprehensive coverage of visual and textual semantics.
Download Link: https://github.com/AwellmanZha/M2ConceptBase
Relevant Paper: https://dl.acm.org/doi/abs/10.1145/3627673.3679852
