Explainable Part-Based Vehicle Classifier with Spatial Awareness
Andreas Caduff, Klaus Zahn, Jonas Hofstetter, Martin Rechsteiner, Patrick Flaig · 2026
Read Explainable Part-Based Vehicle Classifier with Spatial Awareness by Andreas Caduff, Klaus Zahn, Jonas Hofstetter, Martin Rechsteiner, Patrick Flaig (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/arxiv-2605-07831.txt.
Description
Andreas Caduff and four co-authors extend a clear-by-design approach to fine-grained vehicle sorting for smart transport systems. Their earlier design replaced one opaque network with a part detector that finds vehicle parts strong in meaning, followed by feature building and a decision tree. That split made the system easy to extend and partly readable, but treated each part as simply present or absent.
The extension adds spatial awareness: a full probability map conditions each part's presence on the vehicle category being considered, and a softmax regression produces the final probabilities. The change made the classifier much more robust to false part detections, which matters in deployment. Head-to-head tests against a state-of-the-art end-to-end network showed comparable accuracy, challenging the assumed trade-off between accuracy and clarity.
Catalog record
| CREATOR | Andreas Caduff, Klaus Zahn, Jonas Hofstetter, Martin Rechsteiner, Patrick Flaig |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.07831