C3: 90K Paired Floor Plans and Photos with Pixel-Level Correspondences

C3 is a cross-view cross-modality correspondence dataset containing 90,000 paired floor plans and photographs. It covers 597 scenes with 153 million pixel-level correspondences and 85,000 camera poses. The dataset was created by kwhuang and last updated on the platform in January 2026.

Use Cases

• Training models for pixel-level correspondence between architectural plans and real-world photos based on the described paired data.
• Developing algorithms for camera pose estimation based on the 85,000 camera poses mentioned.
• Researching cross-view and cross-modality scene understanding based on the dataset's core design.
• Benchmarking models for tasks like floor plan parsing or image-to-plan matching based on the train/val/test splits.

Strengths

• Contains 90,000 paired data points, providing substantial volume for model training.
• Includes 153 million pixel-level correspondences, offering fine-grained alignment data.
• Covers 597 distinct scenes, suggesting diversity in the visual content.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the main CSV files is unknown, which may limit suitability assessment.
• Data may reflect geographic or source bias inherent to its collection from the Internet.

Provenance

Source

huggingface

Collection Method

Collected from the Internet, as stated in the description.

Time Range

null

Freshness

Last updated 2026-01-05 12:11:34; freshness should be verified.

Geography

null