WildCity: Multi-View Images and HDRI Maps for 3D Reconstruction

WildCity is a dataset from The University of Tokyo designed to train feed-forward 3D Gaussian Splatting models for 3D reconstruction from unconstrained photo collections. It comprises multi-view, multi-lighting images covering 200 scenes and 170 HDRI maps, and includes diverse transient objects. The dataset was last updated on June 11, 2026.

Use Cases

• Training feed-forward 3DGS models based on the described multi-view, multi-lighting image collections.
• Research on 3D reconstruction from unconstrained photo collections based on the dataset's stated purpose.
• Developing models robust to transient objects like people or vehicles based on the dataset's inclusion of diverse transient objects.
• Generating novel views of urban scenes using the provided 200 scenes and 170 HDRI maps.

Strengths

• Contains 200 distinct scenes, providing a substantial base for model training.
• Includes 170 HDRI maps, offering diverse environmental lighting conditions.
• Specifically designed for a cutting-edge task: training feed-forward 3DGS models.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and total file size are unknown, which may limit suitability assessment.
• The dataset's license and specific file formats are not detailed in the provided metadata.

Provenance

Source

Yuto Furutani, Takashi Otonari, Kaede Shiohara, and Toshihiko Yamasaki at The University of Tokyo.

Collection Method

Likely collected as multi-view, multi-lighting photo collections of various scenes.

Freshness

Last updated 2026-06-11 05:25:56; freshness should be verified.