CS639 Computer Vision Final Project Fall 2020

Plant-background image segmentation Marianne Bjorner and Carter Sifferman

Scripts

segment.py

Performs segmentation on an image according to the specified segmentation method. Multiple segmentation methods are defined within the script, and more can be added easily. Written for use on CVPPP2017 dataset.

evaluate_segmentation.py

Tests the accuracy of segmentation using Jaccard score and f1 score. Writes results to a file and saves a histogram of jaccard and f1 score distributions. Written for use on CVPPP2017 dataset, works on any dataset with the same file structure

train_bayes.py

Train a per-pixel naive bayes model to predict whether a pixel belongs to a plant or not. Uses the rgb values of the pixel and trains on each pixel in the given dataset at once. Save trained classifier to naive_bayes.joblib

train_logistic.py

Train a per-pixel logistic regression model to predict whether a pixel belongs to a plant or not. Uses the rgb values of the pixel and trains on each pixel in the given dataset at once. Save trained classifier to logistic.joblib

fix_imgs.py

Fix binary png images so that they're in a consistent binary format. Used for ground truth (_fg) images which can be saved by some image programs as 8 bit RGB pngs rather than binary ones.

tune_slic.py

Tune the parameters of the slic method in segment.py to get the best jaccard score. Results are saved in DATASET_PATH/slic/optimization.txt

tune_logistic_smooth.py

Tune the parameters of the logistic_and_smooth method in segment.py to get the best jaccard score. Results are saved in DATASET_PATH/logistic_and_smooth/optimization.txt

runFluorescentMethod.m

Performs segmentation according to altered MultiLeafTracking method on CVPPP2017 dataset. Original method developed for fluorescent imagery and videos.