Revisiting Reliability in the Reasoning-based Pose Estimation Benchmark

TL;DR

This paper critically examines the reliability of the reasoning-based pose estimation benchmark, identifying key issues and providing refined ground-truth annotations to improve reproducibility and evaluation consistency in pose-aware multimodal large language models.

Contribution

The authors identify reproducibility and quality issues in the RPE benchmark and release refined ground-truth annotations to enhance evaluation fairness and consistency.

Findings

Identified mismatched image indices in the original benchmark

Revealed limitations like image redundancy and scenario imbalance

Provided refined, publicly available ground-truth annotations

Abstract

The reasoning-based pose estimation (RPE) benchmark has emerged as a widely adopted evaluation standard for pose-aware multimodal large language models (MLLMs). Despite its significance, we identified critical reproducibility and benchmark-quality issues that hinder fair and consistent quantitative evaluations. Most notably, the benchmark utilizes different image indices from those of the original 3DPW dataset, forcing researchers into tedious and error-prone manual matching processes to obtain accurate ground-truth (GT) annotations for quantitative metrics (\eg, MPJPE, PA-MPJPE). Furthermore, our analysis reveals several inherent benchmark-quality limitations, including significant image redundancy, scenario imbalance, overly simplistic poses, and ambiguous textual descriptions, collectively undermining reliable evaluations across diverse scenarios. To alleviate manual effort and enhance reproducibility, we carefully refined the GT annotations through meticulous visual matching and publicly release these refined annotations as an open-source resource, thereby promoting consistent quantitative evaluations and facilitating future advancements in human pose-aware multimodal reasoning.