Experiments

_images/experiments.png

Note

This functionality is for now only available for pixelwise (segmentation) projects.

The Experiments tab allows you to easily run your own algorithm on the full dataset. The result of the algorithm on each dataset will be automatically stored in a sub-folder of the project and can be easily visualized later.

In order to learn how to write Python algorithms for ImFusion Labels, please refer to the Writing Algorithms page.

When the experiment has been completed, it is added to the Experiments list. Selecting an experiment updates the Results table on the left side of the window. This table contains all datasets that the experiment has been run on, and shows a preview of the result compared to the ground truth. The accuracy of the experiment is also computed so that you can easily spot the failed cases. The experiments table can be sorted by the different columns by clicking on the header.

When a dataset is selected, it is being automatically loaded in the visualization panel on the right side of the window. Three different visualization of label maps can be selected in the combobox above the views:

  • Ground truth overlays the label map as the user created it,

  • Experiment result overlays the result of the algorithm,

  • Difference overlays in blue the pixels where the ground truth and the algorithm result agree, and in red where they do not match.