Computer Vision News 52 Datasets through the L king-Glass She presented “Metrics Reloaded”, a problem-aware metric recommendation framework resulting from 2.5 years of research involving a total of 73 international experts. The talk emphasizes the notion of “problem fingerprinting” - a structured representation of the biomedical problem capturing all aspects that are relevant for metric selection. Problem fingerprinting consider aspects such as tiny lesions, overlapping targets, uncertainty in the annotations, spatial outliers, etc. that are relevant to your research question. Based on these considerations, the framework recommends suitable metrics for your biomedical problem. There is now available a python library implementation and an interactive online tool; check it out! ③ The speakers of our last talk were Alceu Bissoto and Sandra Avila. Alceu Bissoto is a PhD student at the University of Campinas (Unicamp) and Dr. Sandra Avila is a Professor and Research Scientist at the Institute of Computing, Unicamp, Brazil. First, Sandra presented an overview of their group work on skin lesions over the last 10 years, covering areas such as data augmentation, transfer learning and generative AI. At this point, they found that their skin lesions were so good that they needed to start thinking about bias. Then, Alceu presented a strategy to evaluate bias on
RkJQdWJsaXNoZXIy NTc3NzU=