GeoLabelizer

This project consist in the developpement of a software to help with the visualization and manual labelization/correction of samples based on a set of rasters and a corresponding set of polygons.

It can achieve two different tasks:

• the labelization of a dataset. This mode consists in labelizing from scratch a set of polygons.
• the correction of a dataset. This mode allows to change the value of the predicted class for each sample.

Initialization

Warning: This project was developped on Windows and instances cloned on Linux system have shown difficulties to install some libraries. If possible, it is recommended to install it on a version 3.9 of Python on Windows 11.

There is two ways of setting the environnement up from the root of the project:

• Pip:
  1. Creation of the environement: python -m venv .venv
  2. Connect to it:
     • on Windows: .\.venv\Scripts\activate
     • on Linux: source .venv/bin/activate
  3. Import the libraries through the requirements.txt file: pip install -r requirements.txt
• Conda:
  1. Simply run the following command: conda env create -f environment.yml
  2. Connect to it: conda activate geo-labelizer

Tutorial

Here is a step-by-step procedure to labelize or correct your first dataset:

0. Before starting, check in src\constants.py, that the given extension correspond to the one of your image files. Are the images given in TIF or do you want to read a VRT based on the TIF ?
1. When launching the software by running main.py, the following window will appear.

2. The first step is to load rasters and polygons from the menu load. When loading the polygons, the following form will appear. There, you are asked to choose the mode between the two following options:

• Labelizer: This mode is used to add a column to the dataset with the categories of your choice. You have the possibility to use a selection column by clicking on the corresponding checkbox. If so, you will need to select between the existing columns and then assign a display value for each different values in this column. The app will not allow a column with more than 6 different values. You will then need to enter the name of the new column under Create column of interest and give each value (which is going to be saved inside the resulting file), as well as the corresponding display value. Here also, you can not create more than 6 different values.
• Correcter: This mode aims at correcting an existing column of the dataset. You will need to select which one under Select column of interest and then provide a display value for each different value of the column. The app will also not allow to choose a column with more than 6 different values.

3. Once the poylgon file and the raster location are given, you will now be able to start working on the dataset:

4. When the job is done, or if you want to save your work in progress, you can press the save tab. By doing so, a folder will be created in the same location as your polygon file with the same name + the suffix -corrected.

5. Inside this folder, you will find 3 files:
   • the new version of the polygon file.
     • In correcter mode, it will be a copy with the modifications.
     • In labelizer mode, only the modified samples will be saved in this file
   • a .csv version of the new polygon file for ease of verification purpose
   • a .txt file with the log of all the modification that have been made.

6. Shortkeys:
   • In order to make the usage of the application more ergonomical, the following actions were mapped to the following keys:
     • Move forward: Right arrow, Key 'D', Spacebar
     • Move backward: Left arrow, key 'A'
     • Save project: Ctrl + S