Commuting Training Action

This module provides a set of tools for generating, saving, validating, and training ground truth masks for cochlea stack analysis. These functionalities are part of the VASCilia plugin in Napari.

Features

1. Create Ground Truth (GT) Masks:

   • Initiates the creation of ground truth annotations.

   • Enables users to paint and annotate regions using the Napari paintbrush tool.

   • Provides step-by-step instructions for annotating, saving, and generating masks.

2. Save Ground Truth Masks:

   • Processes annotations to:

     • Remove small components.

     • Fill holes in labeled regions.

     • Retain only the largest connected component for each label.

   • Saves the cleaned masks in a specified directory.

3. Display Stored Ground Truth Masks:

   • Displays previously saved ground truth masks as a 3D stack in Napari.

   • Ensures easy verification of stored annotations.

4. Copy Ground Truth from Segmentation Results:

   • Copies segmentation results into a new ground truth layer.

   • Allows users to refine pre-segmented masks for further analysis.

5. Move Ground Truth Masks:

   • Moves ground truth masks and their corresponding TIFF files to a user-specified directory.

6. Validate Training Data:

   • Checks the integrity of training and validation datasets: - Ensures matching numbers of TIFF and PNG files. - Validates file naming conventions. - Verifies that masks contain valid labels.

   • Ensures the selected training folder contains only Train and Val subdirectories.

7. Train the Model:

   • Trains a deep learning model using the prepared training dataset.

   • Utilizes a progress dialog to show real-time training progress.

   • Provides an estimated time for completion based on the number of iterations.

Error Handling

• Annotation Errors:

  • Ensure labels are within the range of 1 to 255.

  • Correct any invalid or empty labels in masks.

• Validation Errors:

  • Verify that the Train and Val folders contain equal numbers of TIFF and PNG files.

  • Ensure proper file naming conventions (filename.tif matches filename.png).

• Training Errors:

  • Ensure the output model folder is empty before starting training.

  • Validate the training dataset using Check Training Data.

File Management

• Ground truth masks are saved in: <rootfolder>/<filename_base>/Ground_Truth/

• Valid training and validation data are stored in separate Train and Val directories.

• Trained models are saved in the specified output model path in the config.json

Extending the Functionality

To add or modify functionality, edit the following files:

• commute_training_action.py

—

—

—

—

—