Process images with pre-trained models for ANNA-PALM, using Tensorflow-java backend.
For more details, please checkout the ANNA-PALM project website and also this repository.
- Download the Anet-ImageJ plugin: Anet-ImageJ v0.2.2.
- Copy the .jar file into your ImageJ plugin folder (note: Fiji might not work anymore), or directly drag the .jar file into ImageJ (then restart ImageJ).
- You will find an
A-net
entry inPlugins
of ImageJ menu.
- For the first time, click
Setup A-net
in the ImageJPlugins
menu ("Plugins => A-Net => Setup"), in the dialog, click the download button and wait until you see a list of models. - select a model and click
ok
. - Open your image, notice that your image size must match the model input, different model can have different input image size. By default, built-in models has the input size of 512x512. If your input image is bigger or smaller than that, you will need to crop or pad with zeros manually in ImageJ.
- Click
Run A-net
, select input images, and then clickok
- Wait for a while, you should be able to see the result in a pop image window.
Sample images can be downloaded from here: STORM image 1000 frames (used as figure 4 in the paper), cropped to 512x512, full size version (2560x2560), more sample images can be exported from https://annapalm.pasteur.fr.
A set of pre-trained models are provided, a shortcut to download these models is to use the download
button in the plugin. You can check directly in a folder named anet-models
in your ImageJ folder (next to the plugin
folder). You can also download a zip containing the models manually from here.
Model files for the plugin consist of two files: a tensorflow_model.pb
file for the frozen model, and a config.json
file for the configuration which describe the GUI, input and output of the model.
You can create a new folder (e.g. named "mito_model_v2.3") inside the anet-models
folder in ImageJ. Then place your own model file(tensorflow_model.pb
) generated with the freeze.py
script in ANNA-PALM. To define the GUI and describe the model, you need to manually add a config.json
file next to your model file tensorflow_model.pb
. To simplify the process, you can just copy an existing config.json
file provided by us. In case you have a different input image size, you will need to replace all the 512
into your own size in config.json
.
Once you're done, run Setup A-net
from the menu and you should be able to see your model shown in the model list. Then select it and process images with your own model.
Please cite our paper: Ouyang et al., Nat. Biotechnol. 2018, doi:10.1038/nbt.4106