Skip to content


Run a segmentation on a new dataset

Add a new dataset

  1. Create a folder with the dataset title at ROOT_DATA_FOLDER/EM/. e.g. ROOT_DATA_FOLDER/EM/DATASET_TITLE.
  2. Create a folder titled raw in this new folder.
  3. Move the files (pngs/tiff) to the raw folder.
    1. pngs should be labeled {prefix}{index}.png, with index ranging from 0000 to e.g. 0049 for a dataset with resultion z = 50. Note that indexing starts at 0. The prefix can be empty and should be the same for all files to ensure the sorted order.
    2. a tiff file should use the suffix .tif. A certain filename is not required.
  4. Navigate to the dataset create frontend e.g. http://localhost/main/datasets/create.
  5. Fill in the form and save
    1. choose file type pngseq for multiple png files and tif3d for a single tiff file.
    2. The resulution of the files can be obtained with e.g. FijiImage > Show Info.... Width == x, Height == y and Depth == z. For pngseq, z is also equal to the number of png files.
    3. location = e.g. EM/DATASET_TITLE/raw/.
  6. The dataset should show up in the list and be accessible by the viewer.

Open the dataset in the viewer

  1. Navigate to http://localhost/dash/viewer.
  2. Use Selected dataset to show the new dataset.

Run a segmentation using the correct model

  1. Use Selected model to pick the correct pretrained model. e.g. mito 2D when only mitochondria are a class of interest.
  2. Choose a descriptive segmentation.
  3. Click Start new segmentation. The loading bar should show success after <1min.

View the generated segmentation

  1. Click the Refresh button at Selected segmentation in the Viewer.
  2. Selected the new segmentation with the dropdown.

Access the files of the segmentation

  1. The segmentation should be visible using the frontend e.g. http://localhost/main/segmentations/all. In the edit dialogue, the location of the segmentation is visible. You can navigate to this folder on the network disk e.g. ROOT_DATA_FOLDER/segmentations/{DATASET_TITLE}/{SEGMENTATION_TITLE}/.

Add an external annotation to a dataset

  1. Create a folder at annotations/{DATASET_TITLE}/ with the annotation title e.g. ANNOTATION_TITLE.
  2. Add the annotation files to the folder
  3. Create a new annotation via the frontend e.g. http://localhost/main/annotations/create.
    1. location = e.g. annotations/{DATASET_TITLE}/ANNOTATION_FOLDER
    2. TODO type = ...

To work from a previous annotation, duplicate the folder with a different name or reexport to a different folder from your external annotation tool. Then add it via the frontend as usual.

Fine-tune a model with an annotation

Create a new annotation

Use the annotation tools or add an external annotation.

Fine-tune a model using the created annotation

  1. In Fine-tune model
    1. select the model and annotation
    2. epochs = amount of iterations
    3. choose a new model name e.g. FINETUNED_MODEL_MITO
  2. Click Retrain model. The loading bar should show success after 10min-1h.

Run a segmentation using the fine-tuned model

  1. Click refresh at Run segmentation > selected model.
  2. Select the model FINETUNED_MODEL_MITO, choose a new segmentation name and run.

Now you can view the segmentation as shown in a different tutorial.