Skip to main content

A Universe of
Annotated 3D Objects

Objaverse 1.0 is a Massive Dataset with 800K+ Annotated 3D Objects

Ship in a bottle
Pony Cartoon
2021 Lamborghini Countach LPI 800-4
Bathroom Interior
VR store
Ford Mondeo TAXI
Robot Cat - High poly
New York Hot-Dog Cart
Pizza
Cup of cappuccino
Banana
Bread Roll
Komi San vending machine
Sundae
Hummingbird | Flying
Southern White Rhino
Canon AT-1 Retro Camera
Raspberry Pi Pico
Cybertruck - Fun Size
Military Tactical Backpack
Sofa
Air Jordan 1
American side table
Old Oxford Shoe
Christmas Hat
Umbrella
Jade Sword
Thermos - Hydration Bottle 24OZ
Bike Ardis Verona 28
Dart Set
Peeled Banana
Chess Piece Queen
Baseball Glove
Airpods Max
iPhone 11
Grand Piano
Hydria apothecary vase
Bergenia Flower
The Morning Room
Robbedoes Book
Old retro-futuristic spaceship computer
The Great Drawing Room
Headphone with stand
Hurdy-Gurdy
Autotransformer voltage regulator
Monstera Deliciosa Potted Mid-Century plant
Coffee Grinder
1991.45 Table and Tea Service
Jukebox
Hologram Console
Medieval bookcase

Annotations. Each object comes with a text name, description, tags, and additional metadata.

Visual Comparison. A visual comparison of objects in Objaverse vs. ShapeNet.

Car
Bed
Vase
Bag

Scale Comparison. Objaverse is an order of magnitude larger than all previous 3D object datasets.

Animated Objects. Objaverse includes over 44K animated 3D objects.

Visual Diversity. Objaverse includes animated objects, rigged (body-part annotated) characters, models separatable into parts, exterior environments, interior environments, and a wide range visual styles.

Experiments. We have used Objaverse for generating 3D models, as augmentation for 2D instance segmentation, open vocabulary embodied AI, and to study CLIP robustness. We are excited to see all the creative ways that the community will use it!

Download. Objaverse is now openly available and hosted on 🤗 Hugging Face! We are excited to see what you build!

Get started!

Check out the Google Colab and 🤗 Hugging Face repo to get started.

Team. Objaverse was created by the PRIOR team at the Allen Institute for AI and at the Allen School at the University of Washington, Seattle.

Matt Deitke
Dustin Schwenk
Jordi Salvador
Luca Weihs
Oscar Michel
Eli VanderBilt
Ludwig Schmidt
Kiana Ehsani
Ani Kembhavi
Ali Farhadi