Revisiting Reliability in the Reasoning-based Pose Estimation Benchmark

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kim, Junsu, Kim, Naeun, Lee, Jaeho, Park, Incheol, Han, Dongyoon, Baek, Seungryul
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909693498097664
author Kim, Junsu
Kim, Naeun
Lee, Jaeho
Park, Incheol
Han, Dongyoon
Baek, Seungryul
author_facet Kim, Junsu
Kim, Naeun
Lee, Jaeho
Park, Incheol
Han, Dongyoon
Baek, Seungryul
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2507_13314
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Revisiting Reliability in the Reasoning-based Pose Estimation Benchmark
Kim, Junsu
Kim, Naeun
Lee, Jaeho
Park, Incheol
Han, Dongyoon
Baek, Seungryul
Computer Vision and Pattern Recognition
Artificial Intelligence
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.
title Revisiting Reliability in the Reasoning-based Pose Estimation Benchmark
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2507.13314