Architectural Images With AI-Generated Captions

Pexels provided over 10,000 photographs of buildings and unique architecture in 2023. The dataset creator 'lodestones' used the CogVLM model to generate descriptive captions for each image. The dataset was last updated on the Hugging Face platform in June 2024.

Use Cases

• Train an image captioning model using the AI-generated captions from CogVLM as labels for the architectural photographs.
• Fine-tune a vision-language model for architectural style recognition by analyzing the relationship between image content and its corresponding caption.
• Develop a building classification system by using the image filenames or parquet table values as potential metadata for model training.
• Create a retrieval-augmented generation system for architecture by indexing the image-caption pairs for semantic search.

Strengths

• Contains over 10,000 high-resolution photographs sourced from Pexels.
• Each image has a corresponding descriptive caption generated by the CogVLM AI model.
• Data is provided in full image format alongside a structured parquet table for metadata.

Limitations

• The dataset may contain copyrighted building images, introducing legal uncertainty for commercial model deployment.
• Captions are AI-generated, not human-verified, which may introduce label noise or inaccuracies.
• Sample size is moderate (~10k images) and may not cover all architectural styles or regions comprehensively.

Provenance

Source

Pexels (stock photography website).

Collection Method

Images were pulled from Pexels; captions were automatically generated using the CogVLM model.

Time Range

Images sourced in 2023.

Freshness

Data was pulled from Pexels in 2023 and the dataset was last updated on the platform in June 2024.

Geography

null