Our Technology Stack
State-of-the-art, trusted and versatile
Object Detection using Convolutional Neural Networks
Image and Video Anonymization with Deep Learning
Our core technology is object detection using deep convolutional networks (CNN), which in recent years has rapidly evolved from an academic playground to production-ready tools for solving industrial use cases fast and effectively.
We combine different techniques such as instance segmentation and keypoint detection (based on Mask R-CNN or Faster R-CNN with FPN and ResNet backbones) to anonymize people, vehicles, faces or license plates.
We are currently looking into different technologies:
- Semi-supervised methods, e.g. MixMatch, to efficiently train better models using larger datasets,
- Different Generative Adversarial Networks (GAN) to anonymize images by repainting critical areas, which in contrast to blurring minimizes the aesthetic impact while protecting the privacy of the affected people, and
- Lossless Model Compression to further reduce costs and energy consumption, as well as increase portability of our ML models.
For more about our Research & Development and hands-on machine learning tips, check out our Blog.
As part of the NVIDIA Inception program for AI startups, we have access to support, expertise and training by NVIDIA, the leading AI hardware manufacturer.
-  He, Kaiming et al. (2018). Mask R-CNN.
-  Lin, Tsung-Yi et al. (2017). Feature Pyramid Networks for Object Detection.
-  Ren, Shaoqing et al. (2016). Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks.
-  Berthelot, David et al. (2019). MixMatch: A Holistic Approach to Semi-Supervised Learning.
-  Goodfellow, Ian et al. (2014). Generative Adversarial Nets.
-  He, Kaiming et al. (2016). Deep residual learning for image recognition.
Cloud, Web & Software Development
Scalable and Flexible Anonymization Pipeline
While deep learning enables us to automate high-quality anonymization with superior precision and sensitivity, we rely on traditional software engineering craftsmanship to deliver our solution to our customers in a fast and reliable manner.
In collaboration with the AWS Austria team and AWS Activate, we have set up our scalable and flexible cloud system in GDPR-compliant data centres around Europe. Expensive accelerated computing instances are only utilised when required, allowing us to process 200.000+ images per day cost-effectively, to our customer's benefit.
We're looking for you!
If you are interested in creating state-of-the-art ML models and deploying them in a highly scalable and higly available cloud environment, drop us an email and have a chat with us!