DwarfClass

DwarfClass is a specialized tool for generating expert labels of dwarf galaxies. It provides a structured classification interface that can be used to create training datasets for deep learning models.

Features

• Sequential classification workflow with three main questions:

  1. Dwarf galaxy identification (Yes/Maybe/No)
  2. Morphology classification (for "Maybe" or "Yes" responses)
  3. Special features identification (defaults to "No" for efficiency)

• User-friendly interface features:

  • You can use your mouse/touchpad or the keyboard (1,2,3,4) to classify
  • Automatic panel progression and highlighting
  • Previous panels are disabled to maintain classification flow
  • Reset option using ESC key for correcting mistakes
  • Comment box for additional observations
  • Flexible input dynamic:
    • Option 1: use classify_multiple_views_random_order.py with Enter key confirmation at the end of the classifications
    • Option 2: use classify_multiple_views_random_order_v2.py with automatic progression to the next image
  • Randomized image order to prevent bias
  • Line number indicating object position is printed as a mechanic to correct errors once the classification was already locked in

• Progress persistence:

  • Classifications are automatically saved
  • Application can be closed and resumed later
  • Tracks completed classifications and continues with remaining images
  • Saves results in CSV format

Installation

Clone the repository:

git clone https://github.com/heesters-nick/DwarfClass.git
cd DwarfClass

Usage

1. Download the data from this link: https://drive.google.com/drive/folders/1l0brBmMbJbm_FzBgGHQiFv4j8vBUUTgf?usp=sharing

2. Make sure you have all necessary data directories and files. You should have:

   • lsb_gri_prep.h5
   • lsb_gri_prep_binned2x2.h5
   • lsb_gri_prep_binned_smoothed.h5
   • lsb_r_binned2x2.h5
   • train_cutouts_legacy/
   • train_cutouts_legacy_enh/

3. Update local data paths in the script classify_multiple_views_random_order.py under if __name__ == '__main__'::

   • h5_data_dir (should contain all of your files ening in .h5)
   • legacy_data_dir (should contain train_cutouts_legacy and train_cutouts_legacy_enh subdirectories)

4. Start the application:

• Version 1 (confirm by Enter): python classify_multiple_views_random_order.py
• Version 2 (automatic progression): python classify_multiple_views_random_order_v2.py
• You may have to tune the parameter fraction_img_grid such that both the image grid and the button panels fit on the screen. The parameters controls the fraction of vertical space reserved for the image grid.

5. Classification Process:

   • Answer whether the image shows a dwarf galaxy (Yes/Maybe/No)
   • If "Yes" or "Maybe", classify the morphology (dE, dEN, dI, dIN)
   • Add optional comments in the comment box
   • Identify any special features (Globular Clusters (GCs), interacting/tidally disturbed, defaults to "No")
   • If you're using standard script:
     • Finally, confirm classifications with Enter key
   • If you're using v2 script:
     • Automatically progress to the next image once classification are complete

6. Navigation:

   • If you have made a mistake: press ESC to reset current classification
   • Standard script:
     • Comment box can be acessed at any time during the classification process
   • v2 script:
     • Comment box should be accessed either before starting the classification or before it is complete due to automatic progression
   • Press Tab to get out of comment box

Data Management

• Classifications are saved in classification_results.csv in the main directory
• Important: Keep this file in the main directory until the classification run is complete
• Create a new directory to save your completed classifications
• After completing a run:
  1. Move the file to your completed classifications directory
  2. Rename it to classification_results_[INITIALS]_[RUN NUMBER].csv
     • Example: classification_results_NH_1.csv

Key Features

• Randomized Image Order: Images are presented in random order each time the application starts to prevent classification bias
• Progress Tracking: The application remembers which objects have been classified
• Standard script only:
  • Efficient Workflow: Default "No" for special features allows quick progression through typical cases
  • Flexible Input: Complete control over classification process with ability to correct mistakes

Morphology examples from MATLAS

• Remarks:
  • There are only very few examples for nucleated dwarf irregulars (dIN) in MATLAS
  • Some morphologies may seem different in MATLAS and UNIONS due to varying observing conditions
  • In general: if the galaxy appears smooth -> dE, if there are any features/star forming regions -> dI
  • The colors in the Legacy survey images should not be relied upon for morphological classification; all dwarfs look blue in these images