Skip to content

Application for ground-truthing semantic segmentation datasets in PyQt4/OpenCV.

License

Notifications You must be signed in to change notification settings

AngusG/truth-and-crop

Repository files navigation

truth-and-crop

Convenient GUI application for quickly ground-truthing semantic segmentation datasets in Python/OpenCV. Original dataset used with the tool can be downloaded from: https://dataverse.scholarsportal.info/dataset.xhtml?persistentId=doi:10.5683/SP/NTUOK9

sample

Dependencies

  • python 3.4
  • pyqt 4.x
  • opencv 3.x
  • numpy 1.11.x
  • colorama 0.3
  • natsort=5.0.x
  • scikit-image 0.12.x

If using Anaconda, you can use the provided environment.yml file with conda env create -f environment.yml, which will create a virtual environment tnc-py34.

Usage

source activate tnc-py34
python truth_and_crop.py

Buttons

  • Input File - Browse to image file to load.
  • Output Path - Browse to root folder where output should be saved. Three subfolders are automatically created here.
  • Previous/Next Image - If other images were found in same folder as Input File, you can jump between images with these buttons.
  • Refresh - Discards changes.
  • Crop - Switch between annotation mode and cropping mode.
  • Toggle - Toggle annotations on and off to make it easier to see raw image. SLIC is only run on the image for the first toggle, subsequent toggles are much faster.
  • Save - To write all cropped images and masks into appropriate subfolders under the path specified by 'Output Path'.