✅ Just accepted! ✅
📣 We are thrilled to announce that a research paper authored by our PhD students, Yannick Kirchhoff, Maximilian Rokuss, and Saikat Roy has been accepted at #ECCV, a tier-1 research venue (CORE Ranking: A*) in Computer Vision!
❓ What’s the paper about?
Segmenting thin tubular structures like vessels, nerves, roads, or concrete cracks is vital in computer vision but challenging with standard deep learning loss functions. These often compromise structural connectivity, leading to errors in downstream tasks like flow calculation and navigation. While existing topology-focused losses improve accuracy, they are computationally intensive, especially for 3D data and multi-class segmentation.
📝 Their paper, “Skeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular Structures,” addresses this issue by using inexpensive CPU operations, cutting computational overheads by over 90% and excelling in multi-class segmentation. It achieves overall state-of-the-art results on five different datasets and tops the TopCoW and Toothfairy MICCAI challenge leaderboards.
👏 Huge congratulations to Yannick Kirchhoff, Maximilian R., Saikat Roy, and the team: Bálint Kovács, Constantin Ulrich, Tassilo Wald, Maximilian Zenk, supported by seniors Philipp Vollmuth, Jens Kleesiek, Fabian Isensee, and Klaus Maier-Hein!
DKFZ Deutsches Krebsforschungszentrum HIDSS4Health - Helmholtz Information & Data Science School For Health
🔍 Curious about the details? Read the preprint: https://lnkd.in/eCpwQDad.
#eccv #eccv24 #eccv2024 #deeplearning #ai #medicalai #medicalimage #computervision #cvpr #machinelearning