@inproceedings{obrien-etal-2026-evaluation,
title = "Evaluation of Pose Estimation Systems for Sign Language Translation",
author = {O{'}Brien, Catherine and
Sant, Gerard and
M{\"u}ller, Mathias and
Ebling, Sarah},
editor = "Efthimiou, Eleni and
Fotinea, Stavroula-Evita and
Hanke, Thomas and
Hochgesang, Julie A. and
Mesch, Johanna and
Schulder, Marc",
booktitle = "Proceedings of the {LREC} 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion",
month = may,
year = "2026",
address = "Palma, Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.signlang-1.39/",
doi = "10.63317/4shxzirxykmm",
pages = "371--386",
abstract = "Many sign language translation (SLT) systems operate on pose sequences instead of raw video to reduce input dimensionality, improve portability, and partially anonymize signers. The choice of pose estimator is often treated as an implementation detail, with systems defaulting to widely available tools such as MediaPipe Holistic or OpenPose. We present a systematic comparison of pose estimators for pose-based SLT, covering widely used baselines (MediaPipe Holistic, OpenPose) and newer whole-body/high-capacity models (MMPose WholeBody, OpenPifPaf, AlphaPose, SDPose, Sapiens, SMPLest-X). We quantify downstream impact by training a controlled SLT pipeline on RWTH-PHOENIX-Weather 2014 where only the pose representation varies, evaluating with BLEU and BLEURT. To contextualize translation outcomes, we analyze temporal stability, missing hand keypoints, and robustness to occlusion using higher-resolution videos from the Signsuisse dataset. SDPose and Sapiens achieve the best translation performance (BLEU {\textasciitilde}11.5), outperforming the common MediaPipe baseline (BLEU {\textasciitilde}10). In occlusion cases, Sapiens is correct in all tested instances (15/15), while OpenPifPaf fails in nearly all (1/15) and also yields the weakest translation scores. Estimators that frequently leave out hand keypoints are associated with lower BLEU/BLEURT. We release code that can be used not only to reproduce our experiments, but also considerably lowers the barrier for other researchers to use alternative pose estimators."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="obrien-etal-2026-evaluation">
<titleInfo>
<title>Evaluation of Pose Estimation Systems for Sign Language Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Catherine</namePart>
<namePart type="family">O’Brien</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gerard</namePart>
<namePart type="family">Sant</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mathias</namePart>
<namePart type="family">Müller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sarah</namePart>
<namePart type="family">Ebling</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion</title>
</titleInfo>
<name type="personal">
<namePart type="given">Eleni</namePart>
<namePart type="family">Efthimiou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stavroula-Evita</namePart>
<namePart type="family">Fotinea</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="family">Hanke</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Julie</namePart>
<namePart type="given">A</namePart>
<namePart type="family">Hochgesang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Johanna</namePart>
<namePart type="family">Mesch</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marc</namePart>
<namePart type="family">Schulder</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca (Spain)</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Many sign language translation (SLT) systems operate on pose sequences instead of raw video to reduce input dimensionality, improve portability, and partially anonymize signers. The choice of pose estimator is often treated as an implementation detail, with systems defaulting to widely available tools such as MediaPipe Holistic or OpenPose. We present a systematic comparison of pose estimators for pose-based SLT, covering widely used baselines (MediaPipe Holistic, OpenPose) and newer whole-body/high-capacity models (MMPose WholeBody, OpenPifPaf, AlphaPose, SDPose, Sapiens, SMPLest-X). We quantify downstream impact by training a controlled SLT pipeline on RWTH-PHOENIX-Weather 2014 where only the pose representation varies, evaluating with BLEU and BLEURT. To contextualize translation outcomes, we analyze temporal stability, missing hand keypoints, and robustness to occlusion using higher-resolution videos from the Signsuisse dataset. SDPose and Sapiens achieve the best translation performance (BLEU ~11.5), outperforming the common MediaPipe baseline (BLEU ~10). In occlusion cases, Sapiens is correct in all tested instances (15/15), while OpenPifPaf fails in nearly all (1/15) and also yields the weakest translation scores. Estimators that frequently leave out hand keypoints are associated with lower BLEU/BLEURT. We release code that can be used not only to reproduce our experiments, but also considerably lowers the barrier for other researchers to use alternative pose estimators.</abstract>
<identifier type="citekey">obrien-etal-2026-evaluation</identifier>
<identifier type="doi">10.63317/4shxzirxykmm</identifier>
<location>
<url>https://aclanthology.org/2026.signlang-1.39/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>371</start>
<end>386</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Evaluation of Pose Estimation Systems for Sign Language Translation
%A O’Brien, Catherine
%A Sant, Gerard
%A Müller, Mathias
%A Ebling, Sarah
%Y Efthimiou, Eleni
%Y Fotinea, Stavroula-Evita
%Y Hanke, Thomas
%Y Hochgesang, Julie A.
%Y Mesch, Johanna
%Y Schulder, Marc
%S Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F obrien-etal-2026-evaluation
%X Many sign language translation (SLT) systems operate on pose sequences instead of raw video to reduce input dimensionality, improve portability, and partially anonymize signers. The choice of pose estimator is often treated as an implementation detail, with systems defaulting to widely available tools such as MediaPipe Holistic or OpenPose. We present a systematic comparison of pose estimators for pose-based SLT, covering widely used baselines (MediaPipe Holistic, OpenPose) and newer whole-body/high-capacity models (MMPose WholeBody, OpenPifPaf, AlphaPose, SDPose, Sapiens, SMPLest-X). We quantify downstream impact by training a controlled SLT pipeline on RWTH-PHOENIX-Weather 2014 where only the pose representation varies, evaluating with BLEU and BLEURT. To contextualize translation outcomes, we analyze temporal stability, missing hand keypoints, and robustness to occlusion using higher-resolution videos from the Signsuisse dataset. SDPose and Sapiens achieve the best translation performance (BLEU ~11.5), outperforming the common MediaPipe baseline (BLEU ~10). In occlusion cases, Sapiens is correct in all tested instances (15/15), while OpenPifPaf fails in nearly all (1/15) and also yields the weakest translation scores. Estimators that frequently leave out hand keypoints are associated with lower BLEU/BLEURT. We release code that can be used not only to reproduce our experiments, but also considerably lowers the barrier for other researchers to use alternative pose estimators.
%R 10.63317/4shxzirxykmm
%U https://aclanthology.org/2026.signlang-1.39/
%U https://doi.org/10.63317/4shxzirxykmm
%P 371-386
Markdown (Informal)
[Evaluation of Pose Estimation Systems for Sign Language Translation](https://aclanthology.org/2026.signlang-1.39/) (O’Brien et al., SignLang 2026)
ACL
- Catherine O’Brien, Gerard Sant, Mathias Müller, and Sarah Ebling. 2026. Evaluation of Pose Estimation Systems for Sign Language Translation. In Proceedings of the LREC 2026 12th Workshop on the Representation and Processing of Sign Languages: Language in Motion, pages 371–386, Palma, Mallorca (Spain). ELRA Language Resources Association (ELRA).