Fusing Domain-Specific Content from Large Language Models into Knowledge Graphs for Enhanced Zero Shot Object State Classification

TL;DR

This paper explores integrating Large Language Model-generated domain-specific knowledge into knowledge graphs to improve zero-shot object state classification in vision tasks, reducing manual effort and enhancing performance.

Contribution

It introduces a novel pipeline combining LLM-derived semantic embeddings with knowledge graphs for zero-shot classification, demonstrating significant performance gains.

Findings

LLM-based embeddings improve classification accuracy

Combining LLM and pre-trained embeddings yields state-of-the-art results

Extensive ablation study validates the effectiveness of the approach

Abstract

Domain-specific knowledge can significantly contribute to addressing a wide variety of vision tasks. However, the generation of such knowledge entails considerable human labor and time costs. This study investigates the potential of Large Language Models (LLMs) in generating and providing domain-specific information through semantic embeddings. To achieve this, an LLM is integrated into a pipeline that utilizes Knowledge Graphs and pre-trained semantic vectors in the context of the Vision-based Zero-shot Object State Classification task. We thoroughly examine the behavior of the LLM through an extensive ablation study. Our findings reveal that the integration of LLM-based embeddings, in combination with general-purpose pre-trained embeddings, leads to substantial performance improvements. Drawing insights from this ablation study, we conduct a comparative analysis against competing models, thereby highlighting the state-of-the-art performance achieved by the proposed approach.