Solutions
AI Agent Understanding

Indoor smart devices such as robotic vacuum cleaners, home care robots, indoor drones, etc., need to cope with complex indoor environments, and possess the ability for environmental mapping, navigation planning, and object recognition. Coohom Cloud aims to assist smart device manufacturers in cost-effective data collection and utilization through large-scale indoor environmental data, including annotated 2D image datasets and 3D environmental data.

AIGC

In the era of AIGC technology explosion, various generative large-scale models have emerged. Typically, these models are pre-trained based on a large amount of data. Coohom Cloud can provide training data resources including images, videos, models, and other elements, along with a comprehensive labeling system, to assist AIGC researchers in achieving breakthroughs in large-scale model technology.

Learn More
Robotic Simulation

Robot simulation allows for quick, cost-effective, and safe validation of products. Coohom Cloud provides a simulation environment with high rendering quality and physical realism, combined with specific simulation platforms to create a simulation environment database, offering support for enterprise robot simulations.

Learn More
Visualized Product Promotion

The visualized product promotion is to empower customers to experience product charm. Coohom Cloud, through its 3D data resource library and high-fidelity environmental display capabilities, immerse users to in the allure of products within a virtual environment.

Learn More
XR

High-quality datasets significantly enhance the environmental adaptation and user experience of AR/VR/MR systems, such as precise virtual-physical integration effects and smooth interactive operations. Simultaneously, datasets empower content creators to access necessary 3D models, textures, and animation resources, driving content innovation and thus effectively advancing the overall progress and widespread application of XR technology.

Learn More
 中文/EN

Example Datasets

Example datasets to train and evaluate AI in research projects, including 3D models, 3D scenes, and 2D images. Assets are free for evaluation.
Search
All Datasets
Model Datasets
Scene Datasets
Image Datasets
All Datasets
Model Datasets
Scene Datasets
Image Datasets

Model Datasets

All
General Models
Interactive Models
Physical Enhancement Models
OBJ
BLENDER
Electrical Appliances
✓ Basic Information,✓ Material,x Component Structure,x Standard Naming Conventions
FBX
PBR
BLENDER
USD
ProfessionalElectrical Appliances
✓ Basic Information,✓ Material,✓ Component Structure,✓ Standard Naming Conventions
OBJ
BLENDER
Furniture
✓ Basic Information,✓ Material,x Component Structure,x Standard Naming Conventions
FBX
PBR
BLENDER
USD
ProfessionalFurniture
✓ Basic Information,✓ Material,✓ Component Structure,✓ Standard Naming Conventions
OBJ
BLENDER
Teapot
✓ Basic Information,✓ Material,x Component Structure,x Standard Naming Conventions
OBJ
BLENDER
Wardrobe
✓ Basic Information,✓ Material,x Component Structure,x Standard Naming Conventions
USD
BLENDER
Oven
✓ Basic Information,✓ Material,✓ Component Structure,✓ Standard Naming Conventions,✓ Physical Simulation Information: Simulation parameters, collision bodies, parent-child set binding relationships.,✓ Real-world dimensions,✓ Real-world structure
USD
BLENDER
书桌模型
✓ Basic Information,✓ Material,✓ Component Structure,✓ Standard Naming Conventions,✓ Physical Simulation Information: Simulation parameters, collision bodies, parent-child set binding relationships.,✓ Real-world dimensions,✓ Real-world structure
FBX
Flexible deformation
By using physics simulation, we can give more natural and rich postures to common flexible models like socks, pillows, tissues, and so on.
FBX
Splashing deformation
By using physics simulation, irregular stacking can be applied to individual small models like sunflower seeds, peanuts, etc., showcasing a more natural and diverse range of model postures.
FBX
Fractured deformation
By using physics simulation, applying fracture simulation deformation to individual whole models can achieve natural and realistic model fragmentation effects.

Scene Datasets

USD
UE
Indoor Scene
This is a common home decor scene, including dining room, bedroom, bathroom, kitchen, etc. The scene features instance segmentation cleaning, semantic precise labeling, and some models provide component-related information. We used rectangular lighting to simulate realistic lighting in this scene. It is recommended to use Unreal simulation software version greater than 5 or Isaac Sim simulation software version 2022.1 to view this scene.
USD
UE
Commercial Space Scene
This is a typical restaurant scene, including the kitchen, dining hall, bar, etc. The scene features instance segmentation cleaning, precise semantic labeling, and some models provide component-related information. Additionally, we used rectangular lighting to simulate realistic lighting. It is recommended to use Unreal simulation software version greater than 5 or Isaac Sim simulation software version 2022.1.

Image Datasets

OBJ
https://kloudsim-oss.kujiale.com/website_dataset/show_pics/object_recognition/rgb.jpghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/object_recognition/mask.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/object_recognition/depth.png
Object Recognition Dataset
We can select different products and place them in various scenes for recognition. The sample dataset includes two products positioned in four scenes (essentially representing two apartment layouts, each utilizing a utility room and a furnished room).
https://kloudsim-oss.kujiale.com/website_dataset/show_pics/low_I-shaped_traj/rgb.jpghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/low_I-shaped_traj/semantic.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/low_I-shaped_traj/depth.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/low_I-shaped_traj/normal.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/low_I-shaped_traj/texture.png
Indoor Bow-Shaped Trajectory Synthesis Dataset
We have selected an indoor home decor scene to simulate the trajectory of a 150mm radius robotic vacuum cleaner during household cleaning. This trajectory is used to generate camera positions for rendering, with rendering parameters set at a resolution of 1920*1080, FOV60, and a camera height of 50mm. The dataset includes the following types of content: camera poses (intrinsic and extrinsic parameters), depth maps, COCO format 2D image annotation information, normal maps in the camera coordinate system, rendered images, semantic images, and albedo channel images.
https://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_traj/rgb.jpghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_traj/semantic.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_traj/depth.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_traj/normal.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_traj/texture.png
Synthetic Dataset of Indoor Roaming with Random Trajectories
We selected an indoor home decor scene to simulate the perspective of a person walking indoors. The rendering parameters are set as follows: resolution 1920*1080, field of view 60 degrees, camera height 1300mm. The dataset includes the following types of content: camera poses (intrinsic and extrinsic parameters), depth maps, COCO-formatted 2D image annotations, normal maps in the camera coordinate system, rendered images, semantic maps, and albedo channel maps.
https://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_random_traj/rgb.jpghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_random_traj/semantic.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_random_traj/depth.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_random_traj/normal.pnghttps://kloudsim-oss.kujiale.com/website_dataset/show_pics/roam_random_traj/texture.png
Synthetic Dataset of Indoor Captures with Random Height Variation
We selected an indoor home decor scene to simulate roaming rendering with random height and pitch angles. The rendering parameters are set as follows: resolution 1920*1080, field of view 60 degrees, with the camera height randomly varying between 700mm and 1500mm, and the pitch angle randomly varying between -45 degrees and 45 degrees. The dataset includes the following types of content: camera poses (intrinsic and extrinsic parameters), depth maps, COCO-formatted 2D image annotations, normal maps in the camera coordinate system, rendered images, semantic maps, and albedo channel maps.