ConstructionSite 10k: 10,013 Annotated Images for Vision Language Models

ConstructionSite 10k contains 10,013 construction site images and annotations released by LouisChen15 in October 2025. The collection is partitioned into 7,009 training and 3,004 test samples specifically designed to evaluate Vision Language Models (VLMs) in civil engineering contexts.

Use Cases

• Fine-tuning image-to-text models to generate descriptive captions for construction site activities
• Benchmarking Visual Question Answering (VQA) performance on safety-related site queries
• Extracting image features for automated civil engineering site monitoring

Strengths

• 10,013 total images
• Pre-defined 70/30 train-test split
• Domain-specific focus on civil engineering and construction safety

Limitations

• Restricted to non-commercial use via CC BY-NC 4.0 license
• Potential geographic bias toward the United States as indicated by region tags

Provenance

Source

LouisChen15 (Arxiv:2508.11011)

Freshness

Last updated October 2025.

Geography

United States