3D / Point CloudImageMultimodal
GSOT3D
ICCV'25 GSOT3D: Towards Generic 3D Single Object Tracking in the Wild We present a novel benchmark, GSOT3D, that…
<1K·CC-BY-SA·Text (raw)
Objaverse is a large repository of annotated 3D objects released by the Allen Institute for AI (AI2) to support 3D vision, generation, and simulation research. The original Objaverse 1.0 contains roughly 800,000 3D models sourced from Sketchfab, each accompanied by metadata such as titles, descriptions, tags, and creator attribution. The follow-up Objaverse-XL expands this to more than 10 million deduplicated objects gathered from a wider set of sources including GitHub, Thingiverse, and photogrammetry scans.
What’s in it
Assets are provided largely as GLB meshes and range from simple props to detailed, rigged, and textured models spanning everyday items, characters, vehicles, scenes, and scanned artifacts. Many include geometry, materials, textures, and animations, making the collection usable for rendering multi-view images, extracting point clouds, or populating physics simulators. A curated LVIS-annotated subset maps objects to common categories for controlled experiments. Objaverse-XL was a joint effort involving AI2, Columbia, the University of Washington, Stability AI, LAION, and Caltech.
Tasks, license & access
Objaverse is used to train and evaluate 3D generative models (for example Zero123-XL), single-image-to-3D reconstruction, novel-view synthesis, and object recognition, and to supply embodied-AI simulators with diverse assets. The dataset as a whole is distributed under the ODC-By 1.0 license, while individual objects retain their creators’ original licenses; download tooling is provided through a Python API and hosted metadata.
- Modality: 3D meshes (GLB) with textures, materials, metadata, some animations
- Scale: ~800K objects (1.0); 10M+ (Objaverse-XL)
- Tasks: 3D generation, image-to-3D, novel-view synthesis, simulation assets
- License: ODC-By 1.0 (per-object licenses vary)
- Creator: Allen Institute for AI and collaborators
Abundant, diverse 3D geometry is a key bottleneck for spatial and physical AI, so Objaverse matters as one of the largest openly available sources of 3D training data.