An application for quickly ground-truthing semantic segmentation datasets in Python/OpenCV.
python 3.4.5opencv 3.0.0numpy 1.11.2argparse 1.1colorama 0.3.7matplotlib 1.5.3scikit-image 0.12.3
usage: truth_and_crop.py [-h] [--wnd WND] [--ds DS] [--nseg NSEG]
[--sigma SIGMA]
img_path img_name out_path
positional arguments:
img_path path to image to be ground-truthed
img_name name of image to be ground-truthed
out_path root path to save resulting cropped image/mask pairs
optional arguments:
-h, --help show this help message and exit
--wnd WND square crop size / 2 in pixels
--ds DS image down-sampling factor
--nseg NSEG number of superpixel segments for SLIC
--sigma SIGMA gaussian smoothing parameter for SLICExample:
python truth_and_crop.py ~/input_folder/ IMG_6108.JPG ~/output_folder/ --wnd 150 --ds 4 --nseg 200m - Toggle mode between ground-truthing and cropping.w - Write cropped image-mask pairs to IMAGES_OUT_DIR and MASKS_OUT_DIR respectively.q - Quit, note does not automatically save.0 - Select class 0.1 - Select class 1.. so fourth