Computer Vision News - September 2022

40 Best Paper MIUA 2022 Accurate detection and segmentation of polyps is vital in preventing colorectal cancer . In this paper, Javier and Vittorino propose a novel transformer-based architecture for polyp image segmentation called Polyp2Seg . Their method is a hybrid approach, combining the benefits of two architectures: transformers and CNNs . Transformers enable the modeling of long-range dependencies or information between pixels. The CNNs help to extract local information. “ The model has two new blocks – one focused on aggregating information at multiscale levels to characterize polyps of different sizes and one looking at texture and color information, ” Javier explains. Javier Montoya is a Lecturer and Senior Researcher in artificial intelligence in Switzerland. Vittorino Mandujano-Cornejo studied electronic engineering as an undergraduate at the National University of San Marcos in Lima, Peru, and a Master’s in Computer Science at the San Pablo Catholic University in Arequipa. Together, they won the Best Paper award at the recent Medical Image Understanding and Analysis Conference (MIUA) for their work on polyp image segmentation. They are here to tell us more. Javier Montoya Vittorino Mandujano MIUA 2022 Best Paper POLYP2SEG: IMPROVED POLYP SEGMENTATION WITH VISION TRANSFORMER