Megalith-10M-Camera: Predicted Camera Parameters for 9.58 Million Flickr Images

Per-image camera parameter annotations for the Megalith-10M dataset, containing approximately 9.58 million Flickr photos. The annotations were predicted by the Puffin camera-centric multimodal model, which combines a Qwen2.5-7B language model with a C-RADIOv3-H visual encoder. The dataset was created by KangLiao and last updated on June 29, 2026.

Use Cases

• Training or fine-tuning camera parameter estimation models based on the predicted annotations.
• Enhancing image generation or editing models by conditioning on camera metadata.
• Conducting research on the relationship between photographic style and camera settings.
• Augmenting datasets for 3D scene understanding tasks with synthetic camera data.

Strengths

• Large scale with annotations for approximately 9.58 million images.
• Annotations are generated by a sophisticated multimodal AI model (Puffin).
• Data is organized to mirror the source WebDataset layout for consistency.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic, temporal, or stylistic bias inherent to the Flickr source.

Provenance

Source

Flickr photos via the drawthingsai/megalith-10m WebDataset.

Collection Method

Parameters predicted by the Puffin multimodal model.

Freshness

Last updated 2026-06-29 04:27:04; freshness should be verified.