salehahr / graph-extraction-datagen

Generates training data for graph extraction.

graph-training

Generation of training data for the graph extraction of endoscopic images.

Workflow

Video Workflow

1. Append new video data to end of video_data.py
2. Run before_filter.py
3. In MATLAB, run filtering/Bladder_vessels.m (filepath can be input manually)
4. Run after_filter.py

Image Folder Workflow

1. In video_data.py:
   1. Update/Set variable video_filepath = "C:/My/FullPath/ImageFolder" to folder containing the .png images
   2. Set variable use_images = True
   3. If you want to use the FFT filter, set variable fft_filter = True
2. Run before_filter.py
3. In MATLAB, filtering/Bladder_vessels.m:
   1. Update/Set variable line 30 VIDEO_FILEPATH = 'C:/My/FullPath/ImageFolder'
   2. Run filtering/Bladder_vessels.m
4. Run after_filter.py

Functions	Description
before_filter.py	Extracts and crops video frames
filtering/Bladder_vessels.m	Applies B-COSFIRE filter to cropped images
after_filter.py	Applies: mask, thresholding, skeletonising, graph generation

Folders

Folder	Description
raw	Raw video stills
cropped	Cropped images, 256x256px
filtered	Filtered images
masked	Filtered images masked with a circular mask
threshed	Thresholded images
skeleton	Skeletonised images
graphs	Graphs saved as .json files
overlay	Graph overlaid on cropped image