Computer Vision News - April 2022

18 Congrats, Doctor! Christian Marzahl recently completed his PhD at the Pattern Recognition Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg. His research is focused on efficient high-quality and high-quantity dataset annotation and deep learning-based object detection on whole slide images. Furthermore, he conducted studies to analyse how algorithms and experts can efficiently work together and developed the open-source online annotation tool EXACT w hich is already applied in industry, research and academia. Congrats, Doctor Christian! Christian would like to thank his advisors and collaborators AndreasMaier,Marc Aubreville, Christof Bertram and Katharina Breininger for their support during this PhD. Bridging the inter-species gab: Throughout history, humans have learned to treat human diseases by studying animal conditions. I have exploredwhether the same principle can be applied to deep learning. The generalised applicability of deep learning models between species could offer enormous scientific and economic value, especially for domains that lack appropriate training data due to privacy restrictions, data protection or the disease's rarity in certain species. For this purpose, species-independent cell detectionmethods were developed and analysed inmy dissertation, allowing the automatic quantification of pulmonary haemorrhage [1 ] and asthma [ 2 ] on cytological image data. The basis for developing thesemethods is qualitatively and quantitatively comprehensive data sets created online and interdisciplinary with the annotation tool EXACT I developed in my thesis. To support research in inter-domain fields, I co-created and published the first fully annotated multi-species cytopathological pulmonary haemorrhage dataset [ 3 ]. This was made possible by working with an interdisciplinary team to efficiently create this novel data set by combining expert-algorithm collaboration and EXACT in amulti-step pipeline. Initially, an object detectionmodel pre-trained on equine cytology whole slide images was applied to