DiskMask: Focusing Object Features for Accurate Instance Segmentation of Elongated or Overlapping Objects
IEEE International Symposium on Biomedical Imaging (ISBI), 2020
Abstract: Deep learning has enabled automated segmentation in a large variety of cases. Instance segmentation of touching and overlapping objects remains an open challenge. We present an end-to-end approach that focuses object detections and features to local regions in an encoder stage and derives accurate instance masks in a decoder. We avoid heavy pre- or post-processing, such as lifting or non-maximum suppression. The approach compares favorably to the current state-of-the-art on three challenging biological datasets.
Paper
Supplementary
Publisher's link
Images and movies
![teaser.png [614KB]](/Publications/2020/BMB20/teaser.png "teaser.png [614KB]"){kind=small}
See also
- Slides (oral presentation at ISBI 2020, 3-7 April 2020, virtual conference):
- Additional experiments and method extensions (uncertainty estimation, panoptic segmentation and tracking):
BibTex reference
@InProceedings{BMB20,
author = "A. B{\"o}hm and N. Mayer and T. Brox",
title = "DiskMask: Focusing Object Features for Accurate Instance Segmentation of Elongated or Overlapping Objects",
booktitle = "IEEE International Symposium on Biomedical Imaging (ISBI)",
month = " ",
year = "2020",
url = "http://lmb.informatik.uni-freiburg.de/Publications/2020/BMB20"
}