Spotlight
A Probabilistic U-Net for Segmentation of Ambiguous Images
Simon Kohl · Bernardino Romera-Paredes · Clemens Meyer · Jeffrey De Fauw · Joseph R. Ledsam · Klaus Maier-Hein · S. M. Ali Eslami · Danilo Jimenez Rezende · Olaf Ronneberger

Wed Dec 5th 10:35 -- 10:40 AM @ Room 220 E
in Wed Track 2 -- Session 1 »

Many real-world vision problems suffer from inherent ambiguities. In clinical applications for example, it might not be clear from a CT scan alone which particular region is cancer tissue. Therefore a group of graders typically produces a set of diverse but plausible segmentations. We consider the task of learning a distribution over segmentations given an input. To this end we propose a generative segmentation model based on a combination of a U-Net with a conditional variational autoencoder that is capable of efficiently producing an unlimited number of plausible hypotheses. We show on a lung abnormalities segmentation task and on a Cityscapes segmentation task that our model reproduces the possible segmentation variants as well as the frequencies with which they occur, doing so significantly better than published approaches. These models could have a high impact in real-world applications, such as being used as clinical decision-making algorithms accounting for multiple plausible semantic segmentation hypotheses to provide possible diagnoses and recommend further actions to resolve the present ambiguities.

Author Information

Simon Kohl (German Cancer Research Center (DKFZ))
Bernardino Romera-Paredes (DeepMind)
Clemens Meyer (DeepMind)
Jeffrey De Fauw (DeepMind)
Joseph R. Ledsam (DeepMind)
Klaus Maier-Hein (German Cancer Research Center)
Ali Eslami (DeepMind)
Danilo Jimenez Rezende (Google DeepMind)
Olaf Ronneberger (DeepMind)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors