Root Identification in Minirhizotron Imagery with Multiple Instance Learning
Guohao Yu, Alina Zare
If you use this code, cite it: Guohao Yu & Alina Zare. (2019, May 21). GatorSense/MILMinithizotronSegmentation: Initial Release (Version v1.0). Zenodo. http://doi.org/10.5281/zenodo.3122427
In this repository, we provide the code used in paper "Root Identification in Minirhizotron Imagery with Multiple Instance Learning"
This code uses
(1)VLFeat 0.9.21 open source library from matlab
Download the VLFeat binary package "vlfeat-0.9.21-bin.tar.gz" , extracted to the current folder and setup it following https://github.com/vlfeat/vlfeat/blob/master/README.md
(2)LIBSVM library from matlab
Setup the LIBSVM. Run ./libsvm/matlab/make.m
(following https://github.com/cjlin1/libsvm/blob/master/matlab/README)
(3) MIACE
https://github.com/GatorSense/MIACE
case 1: each bag has one inestance or each image per bag
step 1: Run 'Demo_GenearteInstance.m' in MATLAB to generate instance and instance features for MIL algorithm
step 2: Run 'Demo3_mainOneInstanceOneBag.m' in MATLAB to train and test MIL model for the case that there is one instance per bag
or Run 'Demo2_mainOneImageOneBag.m' in MATLAB to train and test MIL model for the case that one image per bag
case 2: there are multiple instances per bag
step 1: Run 'Demo_GenearteInstance.m' in MATLAB to generate instance and instance features for MIL algorithm
step 2: Run 'Demo_GenerateBag.m' in MATLAB to generate bag for MIL algorithm
step 3: Run 'Demo1_mainMulnstanceOneBag.m' in MATLAB to train and test MIL model for the case that there are multple instances per bag
└── root dir
├── dataset //save your dataset in this folder.
├── libsvm //The LIBSVM library
├── MIACE // The MIACE repo
├── parameterfile //parameter files
├── MIMRF_Paper.pdf //related publication
├── vlfeat-0.9.21 //vlfeat-0.9.21 library
├── Demo_GenerateInstance.m // generate instances and instance features
├── Demo_GenerateBag.m // generate bags for bag smaller than whole image but larger than one instance
├── Demo_mainOneInstanceOneBag.m // train and test MIL method for the case that only one instance per bag
├── Demo_mainOneImageOneBag.m // train and test MIL method for the case that image per bag
├── Demo_mainMulnstanceOneBag.m // train and test MIL method for the case that multiple instances per bag
├── preprocessFiles.m // preprocessing image function
├── superpixelFiles.m // generate instance function
├── computeFeaturesFromSuperpixels.m // compute instance features function
├── mul_instance_one_bag.m // training data for the case that multiple instances per bag
├── one_instance_one_bag.m // training data for the case that only one instance per bag
├── one_image_one_bag.m // training data for the case that image per bag
├── one_instance_one_bag.m // compute instance features function
├── miace_det.m // test MIACE function
├── train_misvm.m // train MISVM function
├── train_svm.m // train SVM function
├── train_random_forest.m // train RF function
├── test_svm.m // test SVM function
├── test_RF.m // test RF function
└── util //utility functions
├── find_ins_in_bag.m //find instances in bag
├── generate_bag.m // create bags.
├── generate_instances_in_bag.m // find instances in bag and bag label
├── bag2instance.m //
├── createFileList.m //
├── readFileFromFolder.m //
└── train_test_split.m //split dataset to train and test
This source code is licensed under the license found in the LICENSE file in the root directory of this source tree.
This product is Copyright (c) 2019 G. Yu and A. Zare. All rights reserved.
If you use this algorithm, please cite the following reference using the following BibTeX entries.
@inproceedings{Yu2020RootIdentificationMIL,
title={Root Identification in Minirhizotron Imagery with Multiple Instance Learning},
author={Guohao Yu and Alina Zare and Hudanyun Sheng and Roser Matamala and Joel Reyes-Cabrera and Felix B. Fritschi and Thomas E. Juenger},
journal = {Machine Vision and Applications},
volume = {31},
year={2020}
}