CSBDeep Fiji Plugin
Install
ImageJ update site
The CSBDeep plugin can be installed from the ImageJ update site http://sites.imagej.net/CSBDeep/
. See the CSBDeep Wiki Pages for more details.
From source
- Clone this repository.
- Run the following command from inside the repo:
mvn -Dimagej.app.directory=/path/to/Fiji.app/ -Ddelete.other.versions=true
Run demos
- Download the examplary image data
- Open Fiji.
- Open an example image, e.g.
tribolium.tif
. - Run the plugin via
Plugins > CSBDeep > Demo
. - Run the plugin by pressing
Ok
.
If all goes well, an image will be displayed representing the result of the model execution.
See the CSBDeep Wiki Pages for more details.
Run your own model
- Use the python code to train your network with your data. Export it as ZIP.
- Open Fiji.
- Open an image.
- Run the plugin for any network via
Plugins > CSBDeep > Run your network
. - Load your exported network by pressing
Browse
on theImport model (.zip)
line. - Run the plugin by pressing
Ok
.
If all goes well, an image will be displayed representing the result of the model execution.
See the CSBDeep Wiki Page for more details.
Develop
Code Style
If you use eclipse you can import our code formatter doc/eclipse-code-formatter.xml
, code cleanup (doc/eclipse-code-clean-up.xml
) and import order (eclipse-import-order.importorder
) settings.
Switching the TensorFlow version
For supporting a model trained with a specific TensorFlow version or for GPU support, one has to install a specific native TensorFlow JNI. In order to achieve that, do the following:
- Open
Edit > Options > TensorFlow...
- Choose the version matching your system / model
- Wait until a message opens telling you that the library was installed
- Restart Fiji
GPU support
For GPU support, two things have to be done:
- install CUDA and CuDNN and make sure Fiji knows about the installation paths
- install a TensorFlow version with GPU support (see section above)
For further details please refer to the CSBDeep Wiki page.
Muliple GPUs
See the according CSBDeep Wiki page.
License
This project is licensed under the BSD 2-clause "Simplified" License -- see the LICENSE.txt file for details.