Physics-Based Benchmarking Metrics for Multimodal Synthetic Images
This addresses the need for better evaluation metrics in domain-specific or context-dependent scenarios for researchers and practitioners in multimodal AI.
The paper tackled the problem of evaluating multimodal synthetic images by proposing a new metric called Physics-Constrained Multimodal Data Evaluation (PCMDE), which combines large language models, knowledge mapping, and vision-language models to improve semantic and structural accuracy over existing metrics like BLEU and CLIPScore.
Current state of the art measures like BLEU, CIDEr, VQA score, SigLIP-2 and CLIPScore are often unable to capture semantic or structural accuracy, especially for domain-specific or context-dependent scenarios. For this, this paper proposes a Physics-Constrained Multimodal Data Evaluation (PCMDE) metric combining large language models with reasoning, knowledge based mapping and vision-language models to overcome these limitations. The architecture is comprised of three main stages: (1) feature extraction of spatial and semantic information with multimodal features through object detection and VLMs; (2) Confidence-Weighted Component Fusion for adaptive component-level validation; and (3) physics-guided reasoning using large language models for structural and relational constraints (e.g., alignment, position, consistency) enforcement.