Generative Data Augmentation for ArUco-Free RGB-Based 6-DoF Object Pose Estimation
Carmelo Scribano, Iacopo Ferrari, Giorgia Franchini, Elena Govi, Davide Sapienza, Tobia Poppi, Micaela Verucchi, Marko Bertogna · 2026
Read Generative Data Augmentation for ArUco-Free RGB-Based 6-DoF Object Pose Estimation by Carmelo Scribano, Iacopo Ferrari, Giorgia Franchini, Elena Govi, Davide Sapienza, Tobia Poppi, Micaela Verucchi, Marko Bertogna (2026) 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/pmc-13300756.txt.
Description
Carmelo Scribano and co-authors expose a shortcut bias in a popular pose estimation benchmark in a 2026 computer vision study. The Linemod dataset labels object poses using ArUco fiducial markers on the board, and those markers stay visible in the images. Saliency analysis showed trained networks devoting much of their attention to the markers instead of the objects, inflating their benchmark scores.
The team built a generative augmentation pipeline that erases the markers and swaps in realistic synthesized backgrounds. Models trained on the original data lost noticeable accuracy when judged in marker free scenes, confirming the bias. Retraining on the generative version improved robustness and generalization to unseen settings, although the problem did not vanish completely. The authors urge benchmark designers and industrial users to treat background cues as a first-class threat to reliability.
Catalog record
| CREATOR | Carmelo Scribano, Iacopo Ferrari, Giorgia Franchini, Elena Govi, Davide Sapienza, Tobia Poppi, Micaela Verucchi, Marko Bertogna |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |