Review
Jul 2026
SteelBench: Evaluating Vision-Language Models in Real-World Industrial Environments
STEELBENCH is introduced, a diagnostic benchmark for industrial surveillance that jointly evaluates per-worker activity recognition, safety-rule reasoning, and annotation provenance and finds that unaudited VLM-sourced ground truth can inflate same-family model accuracy by up to 17 percentage points.
Suryanarayana Reddy Yarrabothula, Manisha Chawla, Kunal Sinha et al.
· 1 citation