ShapeSplatsV1: 52,000 3D Objects as Gaussian Splats from ShapeNetCore

ShapeSplatsV1 is a dataset of 52,000 3D objects across 55 categories, derived from the ShapeNetCore repository. The data is distributed as PLY files where each Gaussian splat's information is encoded in custom vertex attributes. The dataset was created by ShapeNet and last updated on Hugging Face in September 2024.

Use Cases

• Train 3D reconstruction models based on Gaussian splat representations.
• Benchmark novel view synthesis algorithms using the 55 object categories.
• Develop generative models for 3D assets based on the large collection of splatted objects.
• Research category-specific 3D shape properties using the labeled object collection.

Strengths

• Contains 52,000 3D objects, providing a substantial scale for training.
• Spans 55 unique object categories from ShapeNetCore, offering categorical diversity.
• Data is derived from established 3D repositories ShapeNetCore, ShapeNet-Part, and ModelNet.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific tasks.
• Last updated 2024-09-05 15:56:05; freshness should be verified.

Provenance

Source

ShapeNet (ShapeNetCore, ShapeNet-Part, ModelNet)

Collection Method

Gathered from existing 3D object repositories and converted to Gaussian splat format.

Freshness

Last updated 2024-09-05 15:56:05.