Contact Us
Want to learn more? We would love to understand your application and discuss ways we can help.
info@artificial-pixels.com
Solutions for your AI products
Does your AI need more data? Could your model perform better?
Our Services
Synthetic Data
Boost your innovation and increase your productivity with synthetic data and system simulations. We can help to start your AI projects earlier and identify needs and issues before running expensive data acquisition campaigns. Whether you want to create pre-trained models or extend your existing datasets with edge cases, we can help to boost your model performance and give you more insights into your problem.
Digital Twins & Custom Addons
Are you considering integrating a digital twin or creating your own workflows to generate synthetic data, enabling your AI team to experiment much faster? With our expertise in computer graphics, we are able to design your digital twin or develop customized add-ons for platforms such as Omniverse or Blender. Give your development team innovative tools so that they can generate high-quality data efficiently and independently.
AI & Machine Vision
Our team of experts with years of experience in the fields can support your projects in many ways. We have been designing, developing, and deploying custom computer vision and AI solutions for years. Our services range from analyzing requirements up to getting tailored solutions for your use case. We successfully bring a project from the conceptualization stage to launch, using sophisticated algorithms and advanced analytics.
Why Synthetic Data?
Data Scarcity: Overcome limitations in real-world data availability by incorporating synthetic data.
Rare Events or Edge Cases: Ensure that AI models can handle infrequent or hard-to-capture scenarios by simulating them.
Expand Diversity: Enhance the diversity of your datasets or simulate near-endless variations for not fully predictable use cases.
Streamline Development: Speed up model iteration cycles by generating data early and rapidly, thus reducing development timelines and costs.
Cost-Effective: Avoid expensive and time-consuming data collection and annotation efforts, saving valuable resources.
How does this work?
Simulation: We employ cutting-edge computer graphics methods to create simulations or digital twins for your use case.
Data Generation: Depending on the application, synthetic data points can be rendered or produced using our controllable generative AI.
Sensor Emulation: The simulation can produce various types of 2D and 3D sensor outputs tailored to match the desired model inputs.
Domain Adaption: To ensure optimal realism for the AI model, the synthetic data can be adjusted to align with the real-world data domain.
Model and Dataset Analysis: When available, existing models and real-world data can be utilized to perform a domain analysis of the problem.
Generative AI
Our generative AI models are suitable for industrial applications, unlike the commonly known and more artistic image generators. That’s because our approach combines generative AI with simulation, allowing complete control over the image content by coupling a digital twin with a generative model. This allows to get the best of both worlds, the capabilities of a generative AI to generate photorealistic images, while maintaining full control over the image content.
Bin Picking
Reusable data generation pipelines are ideal for industrial scenarios. By using procedural scenery elements, physically based simulations, and advanced rendering and post-processing techniques, we can generate unlimited and super realistic training data for your use cases. You can get arbitrary variance in any domain your model needs to generalize, as well as simulations of edge cases to boost your model performance.
Real Footage & Synthetic Data Fusion
With the use of AI, modern Computer Graphics, and VFX techniques also used in the Film Industry, nowadays datasets can be created or extended with near-endless variation and realism for a fraction of the costs of multiple rounds of acquisition and annotation campaigns and with the benefit of reuse- and modifiability. This video shows the fusion of synthetic data with real video footage to demonstrate a possible training data generation use case.
About Us
Dr. Sven Wanner
Sven received his Ph.D. in Physics from Heidelberg University in 2014. His research at the Heidelberg Collaboratory for Image Processing was in the field of 4D light field analysis. His focus was on developing acquisition methods for light fields, as well as 3D- and surface property reconstruction methods in the context of industrial applications and in close collaborations with Bosch and Sony. During his 3 years as postdoc and 6 years working in the industry, he was able to gain experience in product development, as a team leader and worked on multiple different projects, always with a focus on computer vision products, AI, and synthetic data generation.
Dr. Peter Schillen
Peter received his Ph.D. in Scientific Computing and Mathematical Modeling from the University of Mannheim in 2015 after studying Mathematics and Physics at TU Kaiserslautern. He has over 10 years of experience in Image Processing and Deep Learning and has worked on various software and research projects in these areas. As a lead data scientist, he made his contribution to the field of biometric face authentication by creating data pipelines that include data collection hardware, data processing, and continuous metric evaluation. Peter’s passion for technology has helped him contribute to the advancement of emerging technologies in the digital world.