12:00 - 13:00
Modern microscopes can easily produce huge amounts of fascinating image data. Users and their applications are widely spread over various industries and academic research areas, which all have their specific requirements and demands. The talk will give an overview about the ongoing endeavor at Zeiss to develop an open ecosystem for integrated machine-learning workflows, which focusses on the idea of "data-centric" model development, open interfaces, scalable deployment and reproduceable science.
During this journey we created several internal and public python packages, a standardized model format and other tools, which will be showcased. Another aspect are the challenges and learnings that arise from the need to align requirements from industry and academia while keeping our software platform flexible and maintainable, especially when it comes to using Deep Learning methods for image and data processing. Our mission here is to put the scientist or normal user back into the driver seat when it comes to deep learning by creating tools that allow for "data-centric" model development.