This is my graduation project. Based on 3D-Unet, a 3D-segmentation network is implemented.
- test code for data read
- dataloader for pytorch
- generate data for training and validation from raw medical images
- implementation of loss function used in training
- contains some core function, train and predict
- implementaion of metrics
- it is slow, may need to optimize.
- implementation of network model in pytorch
- some auxiliary function
- implement a complete test program using the network
- now finished.
- implement a decorator for time tick.
- implement class for transforms.
- composed of tranforms is supported.
- random select from transforms is supported.
- now working on 3D unet training.
- implement model_zoo.py for different models.
- optimize metrics.py speed.
- transform3d.py for data augumation, enrich methods.
- fix bug in test.py dynamicly select num_classes.
- in np cutter, if stride is too large, may generate minus value in position.
- add remap() in util.py remap [0,1,..7] to origin label value to visualize better.
- multi-resolution enhancement
- multi-model enhancement