README.md 784 Bytes
Newer Older
walker's avatar
walker committed
1
2
3
4
5
6
7
8
9
10
11
12
13
## About
Implementation of 2D Unet for semantic segmentation. 

Unet model is implemented as python package. A tool for ground truth annotation is implemented as python package as well. The training and prediction are run from the jupyter notebooks. See documentation in notebooks for details.

### Features
* Ground truth: supports instance and semantic labelling as input, dense and sparse annotation. Can automatically create a border class.
* Image size: can handle varying image size and does automatic padding.
* Data augmentation: supports augmentation on the fly. Currently implemented: flips
* Postprocessing: In prediction notebook: prediction postprocessing (mask cleanup) and connected component labelling

### Contact
Noreen Walker (Scientific Computing Facility MPI-CBG)