A Deep-Learning Approach to Marble-Burying Quantification: Image Segmentation for Marble and Bedding
Workflow: load the marble/sawdust labels into the marble/sawdust folder in 'Labels'-> load the marble&sawdust labels into the 'Labels-double' folder-> Load the pre-processed inputs to the input folder, make sure each sets were renamed using continuous integers, like '0.tif', '1.tif'
This repository contains the following folders/files
-FilesWithOriginalNaming: the folder with all used images for this project.
-Labels-double: with sub-folder @marbles and @sawdust, in each of the folders, the labels of marbles and sawdusts were listed using the same naming methods, corresponding with the sequential naming of images. The labels were all pre-processed and padded.
-Labels: The labels of marble and sawdust information in a single image, black area as sawdusts, and white areas within the black portion as marbles.
-inputs: Padded, renamed original color images.
-test: A separated folder for small scale testing: for example, testing the prediciton of the model on one image.
-testresult-Yicheng'machine: contains a csv for loss and result data from one test on Yicheng's machine. The xlsx spreadsheet is the overall result generated.
-Convert_combine.m //a matlab script to binarize the two seperate labels and merge them into one png file.
-when changing the k value, also update the variable k here.
-compare.m
-main.py //the main program should be ran after all the original images and labels are in place. uncomment the line 28 and line 88 for kfoldGenerator() if the kfolders were not already generated. (as a mass amount of data, Kfolders were not included in this repository)