CC12M-Camera: Predicted Camera Parameters for 10.9M Images

CC12M-Camera provides per-image camera parameter annotations for the Conceptual 12M dataset. The annotations cover approximately 10.97 million images and were predicted by the Puffin camera-centric multimodal model. The dataset was created by KangLiao and was last updated on June 24, 2026.

Use Cases

• Training camera parameter estimation models based on the predicted annotations.
• Augmenting image datasets with synthetic camera metadata for 3D vision tasks.
• Analyzing the distribution of camera perspectives within a large-scale image collection.
• Fine-tuning multimodal models that require joint understanding of images and camera parameters.

Strengths

• Annotations are provided for a large scale of approximately 10.97 million images.
• The dataset contains 10,902,847 valid annotations, indicating a high validation rate.
• Annotations are generated by a specific, named model (Puffin with Qwen2.5-7B + C-RADIOv3-H).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The annotations are model-predicted and not ground truth, which may introduce systematic errors.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Conceptual 12M (CC12M) dataset

Collection Method

Annotations predicted by the Puffin camera-centric multimodal model.

Freshness

Last updated 2026-06-24 08:53:23; freshness should be verified.