Skip to content

Latest commit

 

History

History
49 lines (35 loc) · 2.84 KB

README.md

File metadata and controls

49 lines (35 loc) · 2.84 KB
Raw

drawing

drawing


Presentation

This repository contains the code of the QGIS plug-in part of DISIR. This is a QGIS plugin designed for interactive semantic segmentation of geo-referenced images using deep-learning.

It is separated into two parts:

  1. frontend deals with the QGIS interface and the interaction with the user
  2. backend deals with a background program launched in a terminal outside QGIS to perform the heavy computations (e.g. the semantic segmentation).

Additional information to better take control of this plugin can be found in this PDF.

Set up

If the backend is used locally, make sure to be in a python environment with the requirements installed.

Install the plugin in QGIS

Compress this repository (qgis_plugin) into a zip file and install the plugin with the Qgis plugin manager.

(Optionnal) Backend in a remote server

This can be useful if the plugin runs on a computer wih a small/no GPU and that a better GPU is accessible remotely.

  • Connect to the remote server using ssh. Check that the local computer hostkey is in the remote server hostkeys: ssh-keygen -H -F <hostname/IP adress> returns some key.
    • If not, add it: ssh-keyscan -H <hostname/IP adress> >> ~/.ssh/known_hosts.
  • In connexion_setup.yml, set the ssh addresses according to your needs:
    • address_server is the IP of the remote server.
    • address_client is the IP of the local computer.
    • username is your username in the local computer.
  • Install the python dependencies in the remote server.

How to start

Enable SSH connexion if the backend is running in a remote server.

  • Launch the backend in a terminal with disir environment activated:
    • Locally: python -m backend
    • In a remote server: python -m backend --ssh
  • In QGIS:
    • Click on th IL button, select your parameters, close it to launch a first inference.
    • Once the first inference is done, click on the red and blue button to select the classes to correct.
    • Click on the misclassified areas
    • Run a new inference by clicking on the NN button

⚠️ The neural network has to take as input a torch tensor of shape C x H x W where $C$ is sorted as follows: [R, G, B, annot_channel_0, ..., annot_channel_n] with n the number of classes.

Tips

Modify the file DISIR/qgis_plugin/frontend/ui/interact_learn_dialog_base.ui using Qt Designer to set the path of the neural network to your own neural network and to set the outuput path to your output directory.