If you have already identified a profound deep learning use case in your organisation, but do not have the expertise to develop a prototype or a scalable and production-ready solution, we can help you out. No matter how far you are in the process, we can support you in all stages of development, from data collection to the final implementation of the solution into regular workflows.
If you have an important problem and sufficient data available, we can develop a tailor-made solution for you - be it prototype or product.
Our hybrid approach leverages on the best of both human intelligence and deep learning. We automate repetitive tasks and leave the most challenging cases to your team. This is more suited for human-centered workflows.
Contrary to our hybrid approach, the autonomous method allows our models to be fully accountable for your data - without any human interaction. This is more suited for real-time cases.
Our work with Daimler was to prove that computer vision techniques are able to check manually picked kits in logistical areas. Within a short period of time and without much data annotation effort, we successfully trained a classification model that is able to analyze the manual kitting processes efficiently and accurately. The model was able to classify objects of different geometric clusters with a high accuracy rate and detect if they are correct or wrong parts. Only a small handful of geometric clusters with indiscernible differences (e.g. a few millimetres difference in height) caused uncertainty in the model.
Infineon designs and produces chips. Chip verification requires lengthy testing to reach functional coverage. We built an unsupervised novelty detection algorithm that reduced redundancy in simulations. Thereby, we saved Infineon Technologies 2,400,000 hours of computational time, saved an amount in the double-digit millions per chip design and decreased time-to-market by 3 months.