About
Reference paper: U-Net: Convolutional Networks for Biomedical Image Segmentation
A brief UNet tensorflow implementation. It can work well on our dataset, see images below. If data augmentation and more strategies are added, the performance will be better.
- You just need to config the config.py to fit your own datast, see Dataset. When the configuration is finished, you can just run and test the model.
- The code will be updated with namescope, tfrecord, and more summaries.
Environment
- Anaconda(python 2.7)
- Tensorflow 1.10
Dataset
The dataset can be organized as follows:
|-- data_path
|-- img_dir_name
|-- annotation_dir_name
|-- train_list_file
|-- trainval_list_file
Train
python train.py
Test
python predict.py