Scene-N1: Matterport3D Assets for Vision-Language Navigation Benchmarks

InternRobotics released Scene-N1 in August 2025, providing a collection of 3D scene assets and Matterport3D scans for Vision-Language Navigation (VLN) research. The repository contains specialized environments for the VLN-N1, VLN-PE, and VLN-CE benchmarks, including residential and commercial scene categories.

Use Cases

• Training navigation agents using the mp3d_n1 base scans for trajectory data
• Evaluating model generalization across internscenes_home and internscenes_commercial environments
• Benchmarking VLN-CE performance using the dedicated mp3d_ce scene assets

Strengths

• Includes specialized assets for three distinct VLN benchmarks (PE, N1, CE)
• Provides both residential and commercial scene categories for evaluation
• Built on the established Matterport3D scan framework with improved materials

Limitations

• Requires specific 3D rendering or simulation environments like Habitat to utilize assets
• Asset-heavy nature implies significant storage requirements

Provenance

Source

InternRobotics and Matterport3D

Collection Method

Derived from Matterport3D scans with custom modifications, material additions, and sky texture augmentations.

Freshness

Last updated August 2025.

Geography

Indoor environments