arXiv:2508.11011v2 Announce Type: replace 
Abstract: Construction safety inspections typically involve a human inspector identifying safety concerns on-site. With the rise of powerful Vision Language Models (VLMs), researchers are exploring their use for tasks such as detecting safety rule violations from on-site images. However, there is a lack of open datasets to comprehensively evaluate and further fine-tune VLMs in construction safety inspection. Current applications of VLMs use small, supervised datasets, limiting their applicability in tasks they are not directly trained for. In this paper, we propose the ConstructionSite 10k, featuring 10,000 construction site images with annotations for three inter-connected tasks, including image captioning, safety rule violation visual question answering (VQA), and construction element visual grounding. Our subsequent evaluation of current state-of-the-art large pre-trained VLMs shows notable generalization abilities in zero-shot and few-shot settings, while additional training is needed to make them applicable to actual construction sites. This dataset allows researchers to train and evaluate their own VLMs with new architectures and techniques, providing a valuable benchmark for construction safety inspection.

قدم الباحثون مجموعة بيانات ConstructionSite 10k، التي تتكون من 10,000 صورة موضوعة من مواقع البناء، بهدف تقييم فعالية نماذج اللغة البصرية الكبيرة المدربة مسبقًا (VLM) في تحديد انتهاكات قواعد السلامة. تتناول هذه المبادرة القيود الحالية في مجموعات البيانات المتاحة لتدريب وتحسين VLM في عمليات التفتيش على السلامة في البناء.

Los investigadores han presentado el conjunto de datos ConstructionSite 10k, que comprende 10,000 imágenes anotadas de sitios de construcción, con el objetivo de evaluar la efectividad de los grandes Modelos de Lenguaje Visual (VLM) preentrenados en la identificación de violaciones de normas de seguridad. Esta iniciativa aborda las limitaciones actuales en los conjuntos de datos disponibles para entrenar y ajustar los VLM en las inspecciones de seguridad de construcción.

Des chercheurs ont introduit le jeu de données ConstructionSite 10k, comprenant 10 000 images annotées de chantiers de construction, visant à évaluer l'efficacité des grands modèles de langage visuel pré-entraînés (VLM) dans l'identification des violations des règles de sécurité. Cette initiative répond aux limitations actuelles des ensembles de données disponibles pour former et affiner les VLM dans les inspections de sécurité en construction.

Researchers have introduced the ConstructionSite 10k dataset, comprising 10,000 annotated images from construction sites, aimed at evaluating the effectiveness of large pre-trained Vision Language Models (VLMs) in identifying safety rule violations. This initiative addresses the current limitations in available datasets for training and fine-tuning VLMs in construction safety inspections.