Scriboora: Rethinking Human Pose Forecasting
Daniel Bermuth, Alexander Poeppel, Wolfgang Reif · 2025
Read Scriboora: Rethinking Human Pose Forecasting by Daniel Bermuth, Alexander Poeppel, Wolfgang Reif (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-15565.txt.
Description
Daniel Bermuth, Alexander Poeppel, and Wolfgang Reif re-examine human pose forecasting, the task of predicting future body poses from past motion. Such forecasts support action recognition, sports analysis, and human-robot interaction. The authors argue progress is hard to judge because studies use varied preprocessing, mismatched scoring code, and only partly shared artifacts.
They audit many recent methods, find several reproducibility problems, and build one shared training and scoring pipeline for absolute pose forecasting. Under that fair protocol, mature sequence models borrowed from speech recognition gave the lowest errors. A paired noisy data set, built from a pose estimator rather than clean motion capture, showed detector noise hurts results badly, though retraining on noisy poses recovers part of the loss.
Catalog record
| CREATOR | Daniel Bermuth, Alexander Poeppel, Wolfgang Reif |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-15565