Reproduce “Multimodal Benchmark for Safety Assessment in Industrial Inspection Scenarios” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zeyi Liu, Shuang Liu, Jihai Min, Zhaoheng Zhang, Jun Cen, Pengyu Han, Songqiao Hu, Zihan Meng and 2 more
InspecSafe-V1 presents a dataset for safety assessment in robot-observed industrial scenes. It pairs RGB imagery and pixel-level object masks with scene descriptions and scenario-dependent safety labels, while additional sensor modalities vary across robots and sites. In an RGB-only VLM benchmark, the reasoning-enhanced Qwen3-VL variant had approximately eight percentage points higher safety-label accuracy than its instruction-tuned counterpart and fewer false positives. This comparison does not test sensor fusion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.