Global probabilistic models for enhancing segmentation with convolutional networks
While deep learning has dramatically improved our capabilities for developing extremely robust segmentation methods, some challenges remain. In many practical settings we only have access to a limited amount of training data. More importantly, the relationship between algorithm performance and the r...
| Main Authors: | , |
|---|---|
| Format: | Conference item |
| Published: |
IEEE
2018
|
| _version_ | 1797070135994875904 |
|---|---|
| author | Fan, M Rittscher, J |
| author_facet | Fan, M Rittscher, J |
| author_sort | Fan, M |
| collection | OXFORD |
| description | While deep learning has dramatically improved our capabilities for developing extremely robust segmentation methods, some challenges remain. In many practical settings we only have access to a limited amount of training data. More importantly, the relationship between algorithm performance and the required amount of training data is not well understood. Here we propose to combine convolutional network based segmentation approaches with a global probabilistic model that effectively enforces prior shape constraints. We demonstrate that the model is capable of accurately segmenting densely packed populations of cells. Our experiments show that combining the convolutional network with the proposed model-based segmentation approach improves the overall segmentation accuracy. |
| first_indexed | 2024-03-06T22:34:45Z |
| format | Conference item |
| id | oxford-uuid:59820caf-9e9c-4524-8b57-d10753d62750 |
| institution | University of Oxford |
| last_indexed | 2024-03-06T22:34:45Z |
| publishDate | 2018 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:59820caf-9e9c-4524-8b57-d10753d627502022-03-26T17:10:07ZGlobal probabilistic models for enhancing segmentation with convolutional networksConference itemhttp://purl.org/coar/resource_type/c_5794uuid:59820caf-9e9c-4524-8b57-d10753d62750Symplectic Elements at OxfordIEEE2018Fan, MRittscher, JWhile deep learning has dramatically improved our capabilities for developing extremely robust segmentation methods, some challenges remain. In many practical settings we only have access to a limited amount of training data. More importantly, the relationship between algorithm performance and the required amount of training data is not well understood. Here we propose to combine convolutional network based segmentation approaches with a global probabilistic model that effectively enforces prior shape constraints. We demonstrate that the model is capable of accurately segmenting densely packed populations of cells. Our experiments show that combining the convolutional network with the proposed model-based segmentation approach improves the overall segmentation accuracy. |
| spellingShingle | Fan, M Rittscher, J Global probabilistic models for enhancing segmentation with convolutional networks |
| title | Global probabilistic models for enhancing segmentation with convolutional networks |
| title_full | Global probabilistic models for enhancing segmentation with convolutional networks |
| title_fullStr | Global probabilistic models for enhancing segmentation with convolutional networks |
| title_full_unstemmed | Global probabilistic models for enhancing segmentation with convolutional networks |
| title_short | Global probabilistic models for enhancing segmentation with convolutional networks |
| title_sort | global probabilistic models for enhancing segmentation with convolutional networks |
| work_keys_str_mv | AT fanm globalprobabilisticmodelsforenhancingsegmentationwithconvolutionalnetworks AT rittscherj globalprobabilisticmodelsforenhancingsegmentationwithconvolutionalnetworks |