Synthetic data. Custom synthetic datasets and simulation scenes for AI training.
It can be. Problems related to transfer results from traning on synthetic data to real world called "sim to real gap". We found that domain randomization, use of hybrid data and specialized domains transferring GANs makes it easier to close it.
Datasets: coco, imagenet
Images: png, jpg
Annotations: json, txt, protobuffs
3d scenes format: .sdf, .usd, .unity, .mujoco
3d models format: .fbx, .obj, .gltf
Work in progress. Coming soon.
Unreal, Unity, PyBullet, Gazebo, Omniverse, MuJoCo.
No. We are not developing new simulation engine. There are plenty already. We are generating simulation scenes (3d playgrounds for your AI agents to play inside) for existing simulators and images/videos datasets.
Sure, we can give you one time free consultation. Describe to us your project and we can talk about what will work best for your needs.
We are developing API to automatically generate datasets.
Currently user facing API is under development, but you can talk to us to describe what you want.
Current workfow is: you contact us -> we will listen to you and ask questions -> we generate you a small dataset sample (10 images or scenes) -> we hear your feedback and adjust -> repeat until you like your sample -> we generate you full datatset.
Yes, but cost and length of the project will probably increase.
Yes, feel free to contact us and describe what you need.
It is reffering to fauvism as an art style.
It depends. Feel free write us an email. We could join based on our workload, recourses availability, cost and how much we understood what your startup is doing.
It depends. Feel free write us an email. We could join based on our workload, recourses availability, cost and how is your cause is close to us.
Yes, we won't provide our service if it is not aligned with our ethical views or could break OFAC regulations. You can't use our service for projects related to military, human rights violations, civilian surveillance, animals rights violations, ecology harm. Message us when in doubts.
Currently not, because of NFT's high carbon footprint. Feel free to message us when it changes.
DALL-E model generates 2d images. We generate labeled and annotated datasets of 2d images and videos and simulation environments.
Those are rendering/game/physics simulation engines. You can use it to run simulation. We don't run the simulations. We are generating virtual environments (simulation scenes, digital playgrounds for your robots to play inside) for all these engines to use.
No. We are not AutoML service. You can use our API to generate datasets and later use it to train models on it on your own.