First steps

After installing the plugin you can open it from the menu under the name ConvPaint. You can use the plugin with various types of images: simple gray-scale, multi-channel, time-lapse, RGB. Note that while you can annotate stacks of images, the learning is never done in real 3D, but all 2D annotations of a stack are are combined for trainig.

To start you can just use one of the images provides as a sample by napari. For example the human mitosis dataset:

Layer selection

If you have multiple image layers open, you can choose which one you want to segment. You can also choose a labels layer to use for annotations. If you don’t yet have a labls layer, you can just click on Add annotations/segmentation layer. Then using the labels tools (on the left of the viewer), you can pick the label to use for drawing. Each label will correspond to a given type of structure you want to segment. If you just want to detect one type of object, you will need two labels: one for background (e.g. 1) and one for objects (2). With the pen tool, you can then create annotations for training:

Train and segment

Once you have added a few annotations, you can train your model. For that just, click on the Train button. A progress bar indicates whether training is finished or not. Once done, you can segment your image with the trained pixel classifier using the Segment image button:

If you are not yet satisfied with the result, you can repeat the cycle: add more annotations, click on Train and click on Segment.

Once you are satisfied, you can save your model to your computer to be able to re-use it later one. For that click on the Save trained model button. This will prompt you to pick a model name and location. To reload the model, click on Load trained model and pick the *.jobib file with your chosen model name. The Current model line indicates what your current model is: no model, a trained but unsaved model or a model with a given name.