Labelling tool

Python based tool for labelling images (here an example for rooftop PV), requires OpenCV be installed (https://pypi.org/project/opencv-python/).

Contacts: N.Varini (EPFL-SCITAS, nicola.varini@epfl.ch) and R.Castello (EPFL-LESO-PB, roberto.castello@epfl.ch)

Usage

Launch the tool by specifying the number of images you want to analyze in one single batch and the directory where images are located

python label_images.py --dir /Users/robertocastello/deneb/labelling_tool/images_from_deneb --batch_size 5

Procedure

• If the image doesn’t contain PV press ’n’ and skip it —> the image will be moved into the /noPV folder

• If the image doeas contain a PV:
  • Crop the area with the mouse functionalities
  • If you make mistakes in the cropping phase, you can restore the original image by pressing 'r'
  • Remember to close the polygon (see mouse commands section)
  • Once you are satisfied, press ‘c’ to visualise the cropped area which will be saved and press 's' to save it
  • The image will be moved into the /PV folder and the labelled one into /label folder with the '_label' suffix

• The scripts terminates automatically after the last image in the batch has been processed

Mouse commands

• Left button: draw lines in between two consecutive clicks
• Central button: close the polygon area
• Right button: erease the last non-closed polygon