Computer Vision News 28 Deep Learning for the Eyes During our conversation, Yannik introduced me to his upcoming publication at MICCAI 2023, titled “Synthesising Rare Cataract Surgery Samples with Guided Diffusion Models”. This work presents promising results in enhancing toolset prediction by synthesizing rare phases as well as tool combinations using the cataract surgery video dataset “CATARACTS”. So, how can image synthesis contribute to a classification task? There are actually two motives. The first one is rather intuitive: reduce the class imbalance by sample generation. Naturally, some surgical phases are shorter than others which commonly results in the network favoring the majority classes. The second objective in their ongoing work is quite interesting: they want to synthesize frames showing human error. While they clearly cannot tell a surgeon to make a mistake on purpose for the sake of a dataset, generating frames for wrong surgery steps seems like a great alternative. A possible application here is training novice surgeons. by Christina Bornberg @datascEYEnce Welcome to the datascEYEnce column! I am Christina and I currently do research in deep learning for ophthalmology at the Singapore Eye Research Institute. I enjoy doing STEM outreach and thanks to Ralph, I can do this now here in Computer Vision News! featuring Yannik Frisch In this edition, I would like to introduce you to the work of Yannik and his colleagues at TU Darmstadt! Yannik is a PhD student at the Interactive Graphics Systems Group focusing on generative models and representation learning applied to surgical data - specifically to the rather neglected field of cataract surgery. Diffusion Models for Sample Synthesis
RkJQdWJsaXNoZXIy NTc3NzU=