Enrich program desscriptions.

Cheng-Lin-Li · Aug 3, 2018 · deff070 · deff070
1 parent 8982050
commit deff070
Show file tree

Hide file tree

Showing 10 changed files with 183 additions and 33 deletions.
diff --git a/README.md b/README.md
@@ -36,7 +36,7 @@ Pre-trained weights include in 'data/saved_models/segcapsr3/split-0_batch-1_shuf
 9. Add 'bce_dice' loss function as binary cross entropy + soft dice coefficient.
 10. Revise 'train', 'test', 'manip' flags from 0 or 1 to flags show up or not to indicate the behavior of main program.
 11. Add new webcam integration program for video stream segmentation.
-
+12. Accept any size of images and program automatically convert to 512 X 512 resolutions.
 ## Procedures 
 
 ### 1. Download this repo to your own folder

diff --git a/gen_mask.py b/gen_mask.py
@@ -11,10 +11,6 @@
 @license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
 
 @contact: clark.cl.li@gmail.com
-@version: 1.2
-
-@create: June 27, 2018
-@updated: July 06, 2018 
 
 Tasks:
 The program implementation will classify input image by a trained model and generate mask image as 

diff --git a/main.py b/main.py
@@ -7,6 +7,37 @@
 
 This is the main file for the project. From here you can train, test, and manipulate the SegCaps of models.
 Please see the README for detailed instructions for this project.
+
+==============
+This is the entry point of the package to train UNet, tiramisu, Capsule Nets (capsbasic) or SegCaps(segcapsr1 or segcapsr3).
+
+@author: Cheng-Lin Li a.k.a. Clark
+
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com
+
+Tasks:
+ The program load parameters for training, testing, manipulation for all models.
+
+
+Data:
+ MS COCO 2017 or LUNA 2016 were tested on this package.
+ You can leverage your own data set but the mask images should follow the format of MS COCO or with background color = 0 on each channel.
+
+Enhancement: 
+ 1. The program was modified to support python 3.6 on Ubuntu 18.04 and Windows 10.
+ 2. Support not only 3D computed tomography scan images but also 2D Microsoft Common Objects in COntext (MS COCO) dataset images.
+ 3. Add Kfold parameter for users to customize the cross validation task. K = 1 will force model to perform overfit.
+ 4. Add retrain parameter to enable users to reload pre-trained weights and retrain the model.
+ 5. Add initial learning rate for users to adjust.
+ 6. Add steps per epoch for users to adjust.
+ 7. Add number of patience for early stop of training to users.
+ 8. Add 'bce_dice' loss function as binary cross entropy + soft dice coefficient.
+ 9. Revise 'train', 'test', 'manip' flags from 0 or 1 to flags show up or not to indicate the behavior of main program.
+ 
 '''
 
 from __future__ import print_function

diff --git a/test.py b/test.py
@@ -6,6 +6,31 @@
 If you have any questions, please email me at lalonde@knights.ucf.edu.
 
 This file is used for testing models. Please see the README for details about testing.
+
+==============
+This is the entry point of the test procedure for UNet, tiramisu, Capsule Nets (capsbasic) or SegCaps(segcapsr1 or segcapsr3).
+
+@author: Cheng-Lin Li a.k.a. Clark
+
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com
+
+Tasks:
+ The program based on parameters from main.py to perform testing tasks on all models.
+
+
+Data:
+ MS COCO 2017 or LUNA 2016 were tested on this package.
+ You can leverage your own data set but the mask images should follow the format of MS COCO or with background color = 0 on each channel.
+ 
+
+Enhancement: 
+ 1. Integrated with MS COCO 2017 dataset.
+ 
+
 '''
 
 from __future__ import print_function

diff --git a/train.py b/train.py
@@ -7,7 +7,29 @@
 
 This file is used for training models. Please see the README for details about training.
 
-Enhancement: Integrated with MS COCO dataset.
+==============
+This is the entry point of the train procedure for UNet, tiramisu, Capsule Nets (capsbasic) or SegCaps(segcapsr1 or segcapsr3).
+
+@author: Cheng-Lin Li a.k.a. Clark
+
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com
+
+Tasks:
+ The program based on parameters from main.py to perform training tasks on all models.
+
+
+Data:
+ MS COCO 2017 or LUNA 2016 were tested on this package.
+ You can leverage your own data set but the mask images should follow the format of MS COCO or with background color = 0 on each channel.
+ 
+
+Enhancement: 
+ 1. Integrated with MS COCO 2017 dataset.
+ 
 '''
 
 from __future__ import print_function

diff --git a/utils/custom_data_aug.py b/utils/custom_data_aug.py
@@ -6,6 +6,27 @@
 If you have any questions, please email me at lalonde@knights.ucf.edu.
 
 This file contains the custom data augmentation functions.
+
+=====
+
+This program includes all data augmentation functions UNet, tiramisu, Capsule Nets (capsbasic) or SegCaps(segcapsr1 or segcapsr3).
+
+@author: Cheng-Lin Li a.k.a. Clark
+
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com 
+ 
+Enhancement:
+ 1. add image2float_array
+ 2. add image_resize2square: Convert any size of image file to 512 X 512 resolutions
+ 3. add image_enhance function to sifht image color space includes background color
+ 4. add process_image to wrap image color shifting and resize the resolutions
+ 5. add change_background_color to change mask background color of MS COCO files to black
+ 6. add convert_mask_data function to resize and change mask color to black
+ 7. add convert_img_data function to wrap process_image, normalized data, reshape to require dimension 3 or 4 
 '''
 
 import numpy as np

diff --git a/utils/custom_losses.py b/utils/custom_losses.py
@@ -7,6 +7,22 @@
 If you have any questions, please email me at lalonde@knights.ucf.edu.
 
 This file contains the definitions of custom loss functions not present in the default Keras.
+
+=====
+
+This program includes all custom loss functions UNet, tiramisu, Capsule Nets (capsbasic) or SegCaps(segcapsr1 or segcapsr3).
+
+@author: Cheng-Lin Li a.k.a. Clark
+
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com 
+ 
+Enhancement:
+ 1. Revise default loss_type to jaccard on dice_soft function. 
+ 2. add bce_dice_loss for future usage.
 '''
 
 import tensorflow as tf

diff --git a/utils/data_helper.py b/utils/data_helper.py
@@ -1,7 +1,18 @@
 '''
+This program is a helper to separate the codes of 2D and 3D image processing.
 
+@author: Cheng-Lin Li a.k.a. Clark
 
-This is a helper file for choosing which dataset functions to create.
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com
+
+Tasks:
+ The program is a helper to separate the codes of 2D and 3D image processing.
+
+ This is a helper file for choosing which dataset functions to create.
 '''
 import logging
 import utils.load_3D_data as ld3D

diff --git a/utils/load_2D_data.py b/utils/load_2D_data.py
@@ -1,18 +1,31 @@
 '''
-Capsules for Object Segmentation (SegCaps)
-Original Paper by Rodney LaLonde and Ulas Bagci (https://arxiv.org/abs/1804.04241)
-Code written by: Rodney LaLonde
-If you use significant portions of this code or the ideas from our paper, please cite it :)
-If you have any questions, please email me at lalonde@knights.ucf.edu.
-
-This file is used for loading training, validation, and testing data into the models.
-It is specifically designed to handle 3D single-channel medical data.
-Modifications will be needed to train/test on normal 3-channel images.
-
-Enhancement:
- 0. Porting to Python version 3.6
- 1. Add image_resize2square to accept any size of images and change to 512 X 512 resolutions.
- 2. 
+This program includes all functions of 2D color image processing for UNet, tiramisu, Capsule Nets (capsbasic) or SegCaps(segcapsr1 or segcapsr3).
+
+@author: Cheng-Lin Li a.k.a. Clark
+
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com
+
+Tasks:
+ The program based on parameters from main.py to load 2D color image files from folders.
+
+ The program will convert all image files into numpy format then store training/testing images into 
+ ./data/np_files and training (and testing) file lists under ./data/split_list folders. 
+ You need to remove these two folders every time if you want to replace your training image and mask files. 
+ The program will only read data from np_files folders.
+ 
+Data:
+ MS COCO 2017 or LUNA 2016 were tested on this package.
+ You can leverage your own data set but the mask images should follow the format of MS COCO or with background color = 0 on each channel.
+ 
+
+Features: 
+ 1. Integrated with MS COCO 2017 dataset.
+ 2. Use PILLOW library instead of SimpleITK for better support on RasberryPi
+ 3. add new generate_test_image function to process single image frame for video stream
 '''
 
 from __future__ import print_function

diff --git a/utils/load_3D_data.py b/utils/load_3D_data.py
@@ -9,34 +9,49 @@
 It is specifically designed to handle 3D single-channel medical data.
 Modifications will be needed to train/test on normal 3-channel images.
 
+=====
+This program includes all functions of 3D image processing for UNet, tiramisu, Capsule Nets (capsbasic) or SegCaps(segcapsr1 or segcapsr3).
+
+@author: Cheng-Lin Li a.k.a. Clark
+
+@copyright: 2018 Cheng-Lin Li@Insight AI. All rights reserved.
+
+@license: Licensed under the Apache License v2.0. http://www.apache.org/licenses/
+
+@contact: clark.cl.li@gmail.com
+
+Tasks:
+ The program based on parameters from main.py to load 3D image files from folders.
+
+ The program will convert all image files into numpy format then store training/testing images into 
+ ./data/np_files and training (and testing) file lists under ./data/split_list folders. 
+ You need to remove these two folders every time if you want to replace your training image and mask files. 
+ The program will only read data from np_files folders.
+ 
+Data:
+ MS COCO 2017 or LUNA 2016 were tested on this package.
+ You can leverage your own data set but the mask images should follow the format of MS COCO or with background color = 0 on each channel.
+ 
+ 
 Enhancement:
- 0. Porting to Python version 3.6
- 1. Add image_resize2square to accept any size of images and change to 512 X 512 resolutions.
- 2. 
+ 1. Porting to Python version 3.6
+ 2. Remove program code cleaning
 '''
 
 from __future__ import print_function
 
 import logging
 from os.path import join, basename
 from os import makedirs
-# from glob import glob
-# import csv
-# from sklearn.model_selection import KFold
+
 import numpy as np
 from numpy.random import rand, shuffle
 import SimpleITK as sitk
-# from sklearn.model_selection import train_test_split
-# from tqdm import tqdm #Progress bar
 
-# import matplotlib
-# matplotlib.use('Agg')
 import matplotlib.pyplot as plt
-# import cv2
 
 plt.ioff()
 
-# from keras.preprocessing.image import *
 from utils.custom_data_aug import augmentImages
 from utils.threadsafe import threadsafe_generator