Physics-Based Benchmarking Metrics for Multimodal Synthetic Images

TL;DR

This paper introduces PCMDE, a physics-constrained multimodal evaluation metric that combines vision-language models and large language models to better assess semantic and structural accuracy in synthetic images.

Contribution

It proposes a novel evaluation framework integrating physics-based reasoning with multimodal feature extraction for improved image assessment.

Findings

PCMDE outperforms traditional metrics in domain-specific scenarios.

The method effectively enforces structural and relational constraints.

It combines object detection, vision-language models, and LLM reasoning for comprehensive evaluation.

Abstract

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.