Skip to main content

Dataset for Robotics

Object Shapes​

  • ShapeNet: ShapeNet is an ongoing effort to establish a richly-annotated, large-scale dataset of 3D shapes. We provide researchers around the world with this data to enable research in computer graphics, computer vision, robotics, and other related disciplines. [Download] [Paper]

  • FineSeg: Shape2Motion paper releases FineSeg dataset, which contains about 3000 3D shapes over six shape categories: chair (1000), table (500), airplanes (600), sofa (600), helicopter (100) and bike (140). The models are collected from a subset of ShapeNet.[Download] [Paper] [Code]

  • PartNet: PartNet is a consistent, large-scale dataset of 3D objects annotated with fine-grained, instance-level, and hierarchical 3D part information. [Download] [Code] [Paper]

  • YCB: YCB Object and Model Set is designed for facilitating benchmarking in robotic manipulation. The set consists of objects in daily life with different shapes, sizes, textures, weight and rigidity, as well as some widely used manipulation tests. [Download]; [Paper]

  • OmniObject3D: OmniObject3D, a large vocabulary 3D object dataset with massive high-quality real-scanned 3D objects to facilitate the development of 3D perception, reconstruction, and generation in the real world. [Download] [Paper] [Code]

  • Objaverse-XL: A Universe of 10M+ 3D Objects. Objaverse-XL is 12x larger than Objaverse 1.0 and 100x larger than all other 3D datasets combined. [Download] [Paper] [Code]

Articulated Objects​

  • PartNet Moblity: SAPIEN releases PartNet-Mobility dataset, which is a collection of 2K articulated objects with motion annotations and rendering material. The dataset powers research for generalizable computer vision and manipulation. [Download] [Paper]

  • AKB48: This is a realistic and physics-rich object repository for articulation analysis. It enables various robotic vision and interaction tasks that require detailed part-level understanding. More categories and download link will be released soon. [Download] [Paper]

  • GAPartNet (Merge and provide more annotations from PartNet Mobility and AKB48): By identifying and defining 9 GAPart classes (lids, handles, etc.) in 27 object categories, we construct a large-scale part-centric interactive dataset, GAPartNet, where we provide rich, part-level annotations (semantics, poses). [Download] [Paper] [Code]

3D Scene Datasets​

DatasetTypeScaleObject / Scene
Annotations		Note
ProcTHOR, ProcTHOR-10kInterior scenes
CAD models
Synthetic		10K
with diverse multi-room layouts
1,633 interactive household objects across 108 categories, with 3,278 materials.		Compatible with AI2-THOR.
Meshes can be extracted via Unity with YandanYang/ai2thor for usage in other simulators.
ReplicaInterior scenes
Scanned meshes		18 in initial release
with a few single-room layouts		3D Sematic segmentation;
Navmesh.
[Aria Ditgital Twin Dataset](https://www.projectaria.com/datasets/adt/)**Interior scenes
Digital twin with CAD models		398 objects (324 stationary, 74 dynamic)
2 real indoor scenes		FPV activities within 200 RGB-D sequences(~400 mins),
with device trajectory, semantic segmentation, 2D/3D bounding boxes, skeleton, and eye gaze).
Aria Synthetic EnvironmentsInterior scenes
CAD models
Synthetic		10K scenes
~8000 3D objects		Simulated cam RGBD with semantics segmentation, and 2D bounding box;
3D floor plan.
HyperSimInterior scenes

CAD models		461 scenes~77K rendered images with
camera trajectory, and 3D bounding boxes.		Built upon some bundles of Evermotion interior data, available here.
Comes with the Hypersim Toolkit
3D FrontInterior scenes

CAD models		18K rooms, 7K furnitureComes with a rendering tool called "Trescope" for 2D rendering with annotations.
Matterport 3D
Habitat-Matterport 3D		Interior scenes
Scanned meshes / CAD models		M3D: 10,800 panoramic views from 90 scanned scenes
HM3D: 1,000 scenes/digital twins.		Semantic and instance segmentation.