Original title: A Benchmark for Vehicle Attribute Classification in Cross-Domain Surveillance Scenarios
Sergio M. Silva, Otavio T. Remer, Gabriel E. Lima, Lucas Wojcik +2 · September 2026
Computer systems track traffic and investigate crimes by analyzing vehicles from camera images. However, models trained in tidy settings struggle in real life because angles, lighting, blockages, and camera types change constantly. To study this problem, researchers built a benchmark called the Unconstrained Vehicle Identification Benchmark, or UVIB. They tested how well models handle three tasks: spotting vehicle orientation, checking if a vehicle is clear enough to identify its make and model, and assessing color clarity. They ran four vision models across different image sources. They discovered that changes in the environment harm model accuracy far more than the choice of computer architecture does. Front and rear orientation transferred well across domains. In contrast, judging make and model suitability suffered from occlusions, while color clarity was very vulnerable to lighting shifts.
Imagine you've practiced identifying cars in a bright, empty parking lot. Now you're at a busy intersection at night with rain and other cars blocking the view. The same cars you knew perfectly now look completely different. In this paper, the benchmark tests how well AI models handle such real-world changes, showing that the difference in conditions matters more than the model used.
Surveillance systems for traffic and crime rely on accurate vehicle recognition. If models fail when conditions change from training to real-world, these systems become unreliable. This benchmark helps measure and improve that robustness.
The abstract does not state that the benchmark covers all possible vehicle types or global surveillance conditions; it only uses images from Brazilian datasets and evaluates four specific architectures.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
A Benchmark for Vehicle Attribute Classification in Cross-Domain Surveillance Scenarios — Sergio M. Silva, Otavio T. Remer, Gabriel E. Lima et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.