Panoinfinigen: Synthetic Panoramic Images with Depth and Normals

PanoInfinigen is a synthetic dataset of high-resolution panoramic images. It was generated using a modified Infinigen framework and the iCity procedural city generator, serving as primary training data for the PaGeR diffusion model. The dataset features perfectly aligned RGB, Depth, and Surface Normal images in ERP format.

Use Cases

• Train single-step diffusion models for zero-shot panoramic depth estimation based on the aligned RGB and depth data.
• Develop and benchmark computer vision algorithms for panoramic scene understanding based on the high-resolution ERP images.
• Generate synthetic urban environments for training autonomous systems based on the urban split created with the iCity generator.
• Research multi-task learning for panoramic geometry based on the aligned RGB, depth, and surface normal features.

Strengths

• Features perfectly aligned RGB, Depth, and Surface Normal images, providing multi-modal ground truth.
• Generated using a modified Infinigen framework specifically for wide-angle panoramic geometry.
• Includes an urban split created with the iCity procedural city generator.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is 2026-05-27 09:41:58.

Provenance

Source

huggingface

Collection Method

Synthetically generated using a modified Infinigen framework and iCity procedural city generator.

Freshness

Last updated 2026-05-27 09:41:58.