Descriptive Captions and Assets for 3D Objects from Multiple Repositories

Cap3D provides descriptive captions for 3D objects from collections including Objaverse, Objaverse-XL, ABO, and ShapeNet. The repository also includes point clouds, rendered images with camera, depth, and MatAlpha information, and Shap-E latent codes for Objaverse objects. Data was created by author 'tiange' for research papers and was last updated in February 2026.

Use Cases

• Train or evaluate 3D captioning models using the descriptive text captions paired with 3D object data.
• Develop view selection algorithms for 3D understanding using the rendered images with associated camera parameters.
• Learn joint embeddings for 3D shapes and text by utilizing the provided Shap-E latent codes alongside their captions.
• Perform 3D reconstruction or analysis tasks using the provided point cloud data for objects.
• Benchmark diffusion-based ranking methods for 3D captioning as referenced in the associated research papers.

Strengths

• Data aggregates 3D objects from four established repositories: Objaverse, Objaverse-XL, ABO, and ShapeNet.
• Includes multiple aligned data modalities per object: captions, point clouds, rendered images, and latent codes.
• Rendered images are accompanied by specific metadata including camera, depth, and MatAlpha information.

Limitations

• Specific scale metrics like total object count, caption count, and dataset size are not provided in the input.
• The temporal coverage and update frequency for the underlying 3D object sources are unknown.
• Potential for class or style imbalance across the four different source collections.

Provenance

Source

Aggregated from Objaverse, Objaverse-XL, ABO, and ShapeNet 3D object repositories.

Collection Method

Created for research on scalable 3D captioning and view selection; includes generated captions, rendered images, and extracted features.

Freshness

Last updated on the platform in February 2026.