OmniLife360: 2,407 Panoramic Sequences for 3D Reconstruction

OmniLife360 is a benchmark for 3D reconstruction from in-the-wild 360-degree captures. It contains 2,407 panoramic sequences totaling about 66 hours, with 1,937 training and 470 testing sequences. The dataset was created by Sg11100 and last updated on 2026-06-26.

Use Cases

• Benchmarking 3D reconstruction models based on in-the-wild 360-degree video sequences.
• Training neural radiance fields (NeRF) or similar methods using panoramic video data.
• Developing novel view synthesis techniques from 360-degree captures.
• Evaluating the robustness of 3D scene understanding algorithms on diverse, real-world environments.

Strengths

• Contains 2,407 panoramic video sequences.
• Provides about 66 hours of total video content.
• Includes a defined split of 1,937 sequences for training and 470 for testing.

Limitations

• The repository is a link-only metadata release; original videos, frames, and audio are not hosted directly.
• Column-level documentation and sample data are unavailable, requiring manual inspection after download.
• The description notes the exclusion of non-essential metadata like EXIF, GPS, and platform context.

Provenance

Source

huggingface

Collection Method

In-the-wild 360-degree captures; specific gathering method is not detailed.

Freshness

Last updated 2026-06-26 09:07:46.