This title appears in the Scientific Report :
2022
Please use the identifier:
http://hdl.handle.net/2128/27736 in citations.
Contour Proposal Networks for Biomedical Instance Segmentation
Contour Proposal Networks for Biomedical Instance Segmentation
We present a conceptually simple framework for object instance segmentation called Contour Proposal Network (CPN), which detects possibly overlapping objects in an image while simultaneously fitting closed object contours using an interpretable, fixed-sized representation based on Fourier Descriptor...
Saved in:
Personal Name(s): | Upschulte, Eric (Corresponding author) |
---|---|
Harmeling, Stefan / Amunts, Katrin / Dickscheid, Timo | |
Contributing Institute: |
Strukturelle und funktionelle Organisation des Gehirns; INM-1 |
Published in: | Medical image analysis (2022) S. 102371 - |
Imprint: |
Amsterdam [u.a.]
Elsevier Science
2021
|
Document Type: |
Preprint |
Research Program: |
Helmholtz Artificial Intelligence Coordination Unit – Local Unit FZJ Human Brain Project Specific Grant Agreement 3 Helmholtz International BigBrain Analytics and Learning Laboratory (HIBALL) JL SMHB - Joint Lab Supercomputing and Modeling for the Human Brain (JL SMHB-2021-2027) Decoding Brain Organization and Dysfunction |
Link: |
Get full text OpenAccess |
Publikationsportal JuSER |
We present a conceptually simple framework for object instance segmentation called Contour Proposal Network (CPN), which detects possibly overlapping objects in an image while simultaneously fitting closed object contours using an interpretable, fixed-sized representation based on Fourier Descriptors. The CPN can incorporate state of the art object detection architectures as backbone networks into a single-stage instance segmentation model that can be trained end-to-end. We construct CPN models with different backbone networks, and apply them to instance segmentation of cells in datasets from different modalities. In our experiments, we show CPNs that outperform U-Nets and Mask R-CNNs in instance segmentation accuracy, and present variants with execution times suitable for real-time applications. The trained models generalize well across different domains of cell types. Since the main assumption of the framework are closed object contours, it is applicable to a wide range of detection problems also outside the biomedical domain. An implementation of the model architecture in PyTorch is freely available. |