Fine-Grained Human Pose Editing Assessment via Layer-Selective MLLMs

2026-01-15

AI Summary

This work addresses structural anomalies and generation artifacts prevalent in existing text-guided human pose editing methods, as well as the absence of fine-grained pose consistency evaluation metrics. To bridge this gap, we propose the first unified evaluation framework based on a layer-selective multimodal large language model (MLLM), integrating contrastive LoRA fine-tuning with Layer Sensitivity Analysis (LSA) to precisely identify optimal feature layers. This enables simultaneous authenticity detection and multidimensional quality regression. Furthermore, we introduce HPE-Bench, a new benchmark dataset designed to support systematic evaluation of pose editing outputs. Extensive experiments demonstrate that our approach achieves state-of-the-art performance on both tasks, effectively bridging the divide between forensic analysis of generated content and comprehensive quality assessment.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-in-the-loop Machine Learning
• Computer Vision: Biometrics, Face, Gesture & Pose

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Text-guided human pose editing has gained significant traction in AIGC applications. However,it remains plagued by structural anomalies and generative artifacts. Existing evaluation metrics often isolate authenticity detection from quality assessment, failing to provide fine-grained insights into pose-specific inconsistencies. To address these limitations, we introduce HPE-Bench, a specialized benchmark comprising 1,700 standardized samples from 17 state-of-the-art editing models, offering both authenticity labels and multi-dimensional quality scores. Furthermore, we propose a unified framework based on layer-selective multimodal large language models (MLLMs). By employing contrastive LoRA tuning and a novel layer sensitivity analysis (LSA) mechanism, we identify the optimal feature layer for pose evaluation. Our framework achieves superior performance in both authenticity detection and multi-dimensional quality regression, effectively bridging the gap between forensic detection and quality assessment.

Problem

• human pose editing
• structural anomalies
• generative artifacts
• evaluation metrics
• pose-specific inconsistencies

Innovation

• layer-selective MLLMs
• HPE-Bench
• layer sensitivity analysis
• contrastive LoRA tuning
• fine-grained pose evaluation