Mind the Shape Gap: A Benchmark and Baseline for Deformation-Aware 6D Pose Estimation of Agricultural Produce

IROS 2026

Nikolas Chatzis*,1,2,3 Angeliki Tsinouka*,1,2 Katerina Papadimitriou1,2,4 Niki Efthymiou1,2 Marios Glytsos2,3,5
George Retsinas2,3 Paris Oikonomou1,2,3 Gerasimos Potamianos1,2,4 Petros Maragos1,2,3 Panagiotis Paraskevas Filntisis1,2
1 HERON – Hellenic Robotics Center of Excellence, Athena Research Center, Greece
2 Institute of Robotics, Athena Research Center, Marousi, Greece
3 School of ECE, National Technical University of Athens, Greece 4 Department of ECE, University of Thessaly, Greece
5 New York University, USA

* Equal contribution

Paper Video Code Dataset soon

Overview

Accurate 6D pose estimation of agricultural produce is challenging because objects from the same category can exhibit substantial instance-level shape variation, creating a mismatch between the observed object and the template geometry used by conventional pose estimators. We introduce PEAR, a real-world benchmark with annotated 6D poses and per-instance 3D shape information, designed both for evaluating 6D pose estimation methods and for systematically studying this shape gap across diverse agricultural produce. Alongside PEAR, we introduce SEED, an RGB-only approach that jointly estimates the 6D pose and deformation of an object from a common category template.

Citation

If you use the video or benchmark, please cite the corresponding paper: 

@inproceedings{Chatzis2026PEAR,
  title={Mind the Shape Gap: A Benchmark and Baseline for Deformation-Aware 6D Pose Estimation of Agricultural Produce},
  author={Nikolas Chatzis and Angeliki Tsinouka and Katerina Papadimitriou and Niki Efthymiou and Marios Glytsos and George Retsinas and Paris Oikonomou and Gerasimos Potamianos and Petros Maragos and Panagiotis Paraskevas Filntisis},
  booktitle={Proceedings of the International Conference on Intelligent Robots and Systems (IROS)},
  year={2026}
}