It is the code for ML class. In the project, we need to find the boundary between different cells (similar to ISBI Challenge[1]). The data set consists of 30 samples, 25 for train and 5 for test given by the teacher. No pretrain or extra data is allowed.
In the repo, we implement three methods, namely:
- UNet
- UNet++
- HRNet
First, you need to put data in the data
directory.
Then you can use the following lines to run the given methods. Examples are given using the default arguments.
UNet:
python train_unet.py
UNet++:
python train_plusplus.py
HRNet:
python train_hrnet.py
Models | IoU | v_info | v_rand |
---|---|---|---|
UNet | 0.679 | 0.916 | 0.886 |
UNet++ | 0.682 | 0.905 | 0.846 |
HRNet | 0.670 | 0.910 | 0.858 |
The repo is originally forked from Pytorch-Unet, but we have changed most part of it.
For convenience and potiential needs, we have kept the original README.md
file, and rename it as README_old.md
.
[1] http://brainiac2.mit.edu/isbi challenge/