3D-Front: 3D Indoor Scenes with Multi-View Images and Point Clouds

A processed version of the 3D-Front dataset, organized into 3D scenes paired with rendered multi-view images and surfaces for the MIDI-3D project. Each scene contains 3D models (.glb), point clouds (.npy), and rendered images in RGB, depth, and normal formats with camera information. The dataset was uploaded by author 'huanngzh' and last updated on 2025-06-25.

Use Cases

• Train 3D scene generation models based on multi-view image and point cloud pairs.
• Develop depth estimation algorithms using the provided depth map renderings.
• Research surface normal prediction from RGB images using the included normal map data.
• Benchmark 3D reconstruction methods using the aligned 3D models and point clouds.

Strengths

• Includes multiple aligned data modalities per scene: 3D models, point clouds, and multi-view images.
• Rendered images provide multiple channels: RGB, depth, and normal maps with camera information.
• Dataset is structured for a specific research project (MIDI-3D), suggesting a defined use case.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Processed from the original 3D-Front dataset.

Collection Method

Data was processed for the MIDI-3D project, involving organization, rendering, and surface extraction.

Freshness

Last updated 2025-06-25 16:38:35; freshness should be verified.