Sumanth ifta hpg

Codes/Deeplab_network/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2017 Zhengyang Wang
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

Codes/Deeplab_network/README.md

@@ -0,0 +1,154 @@
+# Deeplab v2 ResNet for Semantic Image Segmentation
+
+This is an (re-)implementation of [DeepLab v2 (ResNet-101)](http://liangchiehchen.com/projects/DeepLabv2_resnet.html) in TensorFlow for semantic image segmentation on the [PASCAL VOC 2012 dataset](http://host.robots.ox.ac.uk/pascal/VOC/). We refer to [DrSleep's implementation](https://github.com/DrSleep/tensorflow-deeplab-resnet) (Many thanks!). We do not use tf-to-caffe packages like kaffe so you only need TensorFlow 1.3.0+ to run this code.
+
+The deeplab pre-trained ResNet-101 ckpt files (pre-trained on MSCOCO) are provided by DrSleep -- [here](https://drive.google.com/drive/folders/0B_rootXHuswsZ0E4Mjh1ZU5xZVU). Thanks again!
+
+Created by [Zhengyang Wang](http://www.eecs.wsu.edu/~zwang6/) and [Shuiwang Ji](http://www.eecs.wsu.edu/~sji/) at Washington State University.
+
+## Update
+
+**12/13/2017**:
+
+* Now the test code will output the mIoU as well as the IoU for each class.
+
+**12/12/2017**:
+
+* Add 'predict' function, you can use '--option=predict' to save your outputs now (both the true prediction where each pixel is between 0 and 20 and the visual one where each class has its own color).
+
+* Add multi-scale training, testing and predicting. Check main_msc.py and model_msc.py and use them just as main.py and model.py.
+
+* Add plot_training_curve.py to use the log.txt to make plots of training curve.
+
+* Now this is a 'full' (re-)implementation of [DeepLab v2 (ResNet-101)](http://liangchiehchen.com/projects/DeepLabv2_resnet.html) in TensorFlow. Thank you for the support. You are welcome to report your settings and results as well as any bug!
+
+**11/09/2017**:
+
+* The new version enables using original ImageNet pre-trained ResNet models (without pre-training on MSCOCO). You may change arguments ('encoder_name' and 'pretrain_file') in main.py to use corresponding pre-trained models. The original pre-trained ResNet-101 ckpt files are provided by tensorflow officially -- [res101](http://download.tensorflow.org/models/resnet_v1_101_2016_08_28.tar.gz) and [res50](http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz).
+
+* To help those who want to use this model on the CityScapes dataset, I shared the corresponding txt files and the python file which generates them. Note that you need to use tools [here](https://github.com/mcordts/cityscapesScripts) to generate labels with trainID first. Hope it would be helpful. Do not forget to change IMG_MEAN in model.py and other settings in main.py.
+
+* 'is_training' argument is removed and 'self._batch_norm' changes. Basically, for a small batch size, it is better to keep the statistics of the BN layers (running means and variances) frozen, and to not update the values provided by the pre-trained model by setting 'is_training=False'. Note that is_training=False still updates BN parameters gamma (scale) and beta (offset) if they are presented in var_list of the optimiser definition. Set 'trainable=False' in BN fuctions to remove them from trainable_variables.
+
+* Add 'phase' argument in network.py for future development. 'phase=True' means training. It is mainly for controlling batch normalization (if any) in the non-pre-trained part.
+```
+Example: If you have a batch normalization layer in the decoder, you should use 
+
+outputs = self._batch_norm(inputs, name='g_bn1', is_training=self.phase, activation_fn=tf.nn.relu, trainable=True)
+```
+* Some changes to make the code more readable and easy to modify for future research.
+
+* I plan to add 'predict' function to enable saving predicted results for offline evaluation, post-processing, etc.
+
+## System requirement
+
+#### Programming language
+```
+Python 3.5
+```
+#### Python Packages
+```
+tensorflow-gpu 1.3.0
+```
+## Configure the network
+
+All network hyperparameters are configured in main.py.
+
+#### Training
+```
+num_steps: how many iterations to train
+
+save_interval: how many steps to save the model
+
+random_seed: random seed for tensorflow
+
+weight_decay: l2 regularization parameter
+
+learning_rate: initial learning rate
+
+power: parameter for poly learning rate
+
+momentum: momentum
+
+encoder_name: name of pre-trained model, res101, res50 or deeplab
+
+pretrain_file: the initial pre-trained model file for transfer learning
+
+data_list: training data list file
+
+grad_update_every (msc only): accumulate the gradients for how many steps before updating weights. Note that in the msc case, this is actually the true training batch size.
+```
+#### Testing/Validation
+```
+valid_step: checkpoint number for testing/validation
+
+valid_num_steps: = number of testing/validation samples
+
+valid_data_list: testing/validation data list file
+```
+#### Data
+```
+data_dir: data directory
+
+batch_size: training batch size
+
+input height: height of input image
+
+input width: width of input image
+
+num_classes: number of classes
+
+ignore_label: label pixel value that should be ignored
+
+random_scale: whether to perform random scaling data-augmentation
+
+random_mirror: whether to perform random left-right flipping data-augmentation
+```
+#### Log
+```
+modeldir: where to store saved models
+
+logfile: where to store training log
+
+logdir: where to store log for tensorboard
+```
+## Training and Testing
+
+#### Start training
+
+After configuring the network, we can start to train. Run
+```
+python main.py
+```
+The training of Deeplab v2 ResNet will start.
+
+#### Training process visualization
+
+We employ tensorboard for visualization.
+
+```
+tensorboard --logdir=log --port=6006
+```
+
+You may visualize the graph of the model and (training images + groud truth labels + predicted labels).
+
+To visualize the training loss curve, write your own script to make use of the training log.
+
+#### Testing and prediction
+
+Select a checkpoint to test/validate your model in terms of pixel accuracy and mean IoU.
+
+Fill the valid_step in main.py with the checkpoint you want to test. Change valid_num_steps and valid_data_list accordingly. Run
+
+```
+python main.py --option=test
+```
+
+The final output includes pixel accuracy and mean IoU.
+
+Run
+
+```
+python main.py --option=predict
+```
+The outputs will be saved in the 'output' folder.

Codes/Deeplab_network/main.py

@@ -0,0 +1,137 @@
+import argparse
+import os
+import tensorflow as tf
+from model import Model
+
+
+
+"""
+This script defines hyperparameters.
+"""
+
+
+
+def configure(test_data_list_, out_dir_, test_step_, test_num_steps_, modeldir_, data_dir_, num_steps_,
+	save_interval_, learning_rate_, pretrain_file_, data_list_, batch_size_, input_height_, input_width_,
+	num_classes_, print_color_, log_dir_, log_file_,encoder_name_):
+	flags = tf.app.flags
+
+	# training
+	flags.DEFINE_integer('num_steps', num_steps_, 'maximum number of iterations')
+	flags.DEFINE_integer('save_interval', save_interval_, 'number of iterations for saving and visualization')
+	flags.DEFINE_integer('random_seed', 1234, 'random seed')
+	flags.DEFINE_float('weight_decay', 0.0005, 'weight decay rate')
+	flags.DEFINE_float('learning_rate', learning_rate_, 'learning rate')
+	flags.DEFINE_float('power', 0.9, 'hyperparameter for poly learning rate')
+	flags.DEFINE_float('momentum', 0.9, 'momentum')
+	flags.DEFINE_string('encoder_name', encoder_name_, 'name of pre-trained model, res101, res50 or deeplab')
+	flags.DEFINE_string('pretrain_file', pretrain_file_, 'pre-trained model filename corresponding to encoder_name')
+	flags.DEFINE_string('data_list', data_list_, 'training data list filename')
+
+	# validation
+	flags.DEFINE_integer('valid_step', 217000, 'checkpoint number for validation')
+	flags.DEFINE_integer('valid_num_steps', 81605, '= number of validation samples')
+	flags.DEFINE_string('valid_data_list', './dataAugment/val.txt', 'validation data list filename')
+
+	# prediction / saving outputs for testing or validation
+	flags.DEFINE_string('out_dir', out_dir_, 'directory for saving outputs')
+	flags.DEFINE_integer('test_step', test_step_, 'checkpoint number for testing/validation')
+	flags.DEFINE_integer('test_num_steps', test_num_steps_, '= number of testing/validation samples')
+	flags.DEFINE_string('test_data_list', test_data_list_, 'testing/validation data list filename')
+	flags.DEFINE_boolean('visual', False, 'whether to save predictions for visualization')
+
+	# data
+	flags.DEFINE_string('data_dir', data_dir_, 'data directory')
+	flags.DEFINE_integer('batch_size', batch_size_, 'training batch size')
+	flags.DEFINE_integer('input_height', input_height_, 'input image height')
+	flags.DEFINE_integer('input_width', input_width_, 'input image width')
+	flags.DEFINE_integer('num_classes', num_classes_, 'number of classes')
+	flags.DEFINE_integer('ignore_label', 255, 'label pixel value that should be ignored')
+	flags.DEFINE_boolean('random_scale', False, 'whether to perform random scaling data-augmentation')
+	flags.DEFINE_boolean('random_mirror', False, 'whether to perform random left-right flipping data-augmentation')
+
+	# log
+	flags.DEFINE_string('modeldir', modeldir_, 'model directory')
+	flags.DEFINE_string('logfile', log_file_, 'training log filename')
+	flags.DEFINE_string('logdir', log_dir_, 'training log directory')
+
+	# text color
+	flags.DEFINE_string('print_color', print_color_, 'color of printed outputs')
+
+
+
+	flags.FLAGS.__dict__['__parsed'] = False
+	return flags.FLAGS
+
+def main(_):
+	if args.option not in ['train', 'test', 'predict']:
+		print('invalid option: ', args.option)
+		print("Please input a option: train, test, or predict")
+	else:
+		# Set up tf session and initialize variables.
+		# config = tf.ConfigProto()
+		# config.gpu_options.allow_growth = True
+		# sess = tf.Session(config=config)
+		sess = tf.Session()
+		# Run
+		model = Model(sess, configure(test_data_list_=args.test_data_list, out_dir_=args.out_dir, test_step_=args.test_step,
+		test_num_steps_=args.test_num_steps, modeldir_=args.modeldir, data_dir_=args.data_dir,
+		num_steps_=args.num_steps, save_interval_=args.save_interval, learning_rate_=args.learning_rate,
+		pretrain_file_=args.pretrain_file, data_list_=args.data_list, batch_size_=args.batch_size,
+		input_height_=args.input_height, input_width_=args.input_width, num_classes_=args.num_classes,
+		print_color_=args.print_color, log_dir_=args.log_dir, log_file_=args.log_file, encoder_name_=args.encoder_name))
+		getattr(model, args.option)()
+
+
+if __name__ == '__main__':
+	parser = argparse.ArgumentParser()
+
+	parser.add_argument('--option', dest='option', type=str, default='train',
+		help='actions: train, test, or predict')
+	parser.add_argument('--test_data_list', dest='test_data_list', type=str, default='./dataset/test.txt',
+		help='testing/validation data list filename')
+	parser.add_argument('--out_dir', dest='out_dir', type=str, default='output',
+		help='directory for saving testing outputs')
+	parser.add_argument('--test_step', dest='test_step', type=int, default=350000,
+		help='checkpoint number for testing/validation')
+	parser.add_argument('--test_num_steps', dest='test_num_steps', type=int, default=81605,
+		help='number of testing/validation samples')
+	parser.add_argument('--modeldir', dest='modeldir', type=str, default='modelAugment',
+		help='model directory')
+	parser.add_argument('--data_dir', dest='data_dir', type=str, default='/hdd/wsi_fun/ImageAugCustom/AugmentationOutput',
+		help='data directory')
+	parser.add_argument('--gpu', dest='gpu', type=str, default='0',
+		help='specify which GPU to use')
+	parser.add_argument('--num_steps', dest='num_steps', type=int, default=100000,
+		help='maximum number of iterations')
+	parser.add_argument('--save_interval', dest='save_interval', type=int, default=15000,
+		help='number of iterations for saving and visualization')
+	parser.add_argument('--learning_rate', dest='learning_rate', type=float, default=2.5e-4,
+		help='learning rate')
+	parser.add_argument('--pretrain_file', dest='pretrain_file', type=str, default='deeplab_resnet.ckpt',
+		help='pre-trained model filename corresponding to encoder_name')
+	parser.add_argument('--data_list', dest='data_list', type=str, default='./dataAugment/train.txt',
+		help='training data list filename')
+	parser.add_argument('--batch_size', dest='batch_size', type=int, default=15,
+		help='training batch size')
+	parser.add_argument('--input_height', dest='input_height', type=int, default=256,
+		help='input image height')
+	parser.add_argument('--input_width', dest='input_width', type=int, default=256,
+		help='input image width')
+	parser.add_argument('--num_classes', dest='num_classes', type=int, default=2,
+		help='number of classes in images')
+	parser.add_argument('--log_dir', dest='log_dir', type=str, default="log",
+		help='directory for saving log files')
+	parser.add_argument('--log_file', dest='log_file', type=str, default="log.txt",
+		help='Default logfile name')
+	parser.add_argument('--print_color', dest='print_color', type=str, default="\033[0;37;40m",
+		help='color of printed text')
+	parser.add_argument('--encoder_name', dest='encoder_name', type=str, default=" ",
+		help='color of printed text')
+
+
+	args = parser.parse_args()
+
+	# Choose which gpu or cpu to use
+	os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
+	tf.app.run()

Codes/Deeplab_network/main.py~

@@ -0,0 +1,82 @@
+import argparse
+import os
+import tensorflow as tf
+from model import Model
+
+
+
+"""
+This script defines hyperparameters.
+"""
+
+
+
+def configure():
+	flags = tf.app.flags
+
+	# training
+	flags.DEFINE_integer('num_steps', 250000, 'maximum number of iterations')
+	flags.DEFINE_integer('save_interval', 5000, 'number of iterations for saving and visualization')
+	flags.DEFINE_integer('random_seed', 1234, 'random seed')
+	flags.DEFINE_float('weight_decay', 0.0005, 'weight decay rate')
+	flags.DEFINE_float('learning_rate', 2.5e-4, 'learning rate')
+	flags.DEFINE_float('power', 0.9, 'hyperparameter for poly learning rate')
+	flags.DEFINE_float('momentum', 0.9, 'momentum')
+	flags.DEFINE_string('encoder_name', 'deeplab', 'name of pre-trained model, res101, res50 or deeplab')
+	flags.DEFINE_string('pretrain_file', 'model3/', 'pre-trained model filename corresponding to encoder_name')
+	flags.DEFINE_string('data_list', './dataset/train.txt', 'training data list filename')
+
+	# validation
+	flags.DEFINE_integer('valid_step', 200000, 'checkpoint number for validation')
+	flags.DEFINE_integer('valid_num_steps', 32659, '= number of validation samples')
+	flags.DEFINE_string('valid_data_list', './dataset/val.txt', 'validation data list filename')
+
+	# prediction / saving outputs for testing or validation
+	flags.DEFINE_string('out_dir', 'output', 'directory for saving outputs')
+	flags.DEFINE_integer('test_step', 60000, 'checkpoint number for testing/validation')
+	flags.DEFINE_integer('test_num_steps', 48209, '= number of testing/validation samples')
+	flags.DEFINE_string('test_data_list', './dataset/test.txt', 'testing/validation data list filename')
+	flags.DEFINE_boolean('visual', True, 'whether to save predictions for visualization')
+
+	# data
+	flags.DEFINE_string('data_dir', '/hdd/wsi_fun/wsi_data/wsi_blocks/', 'data directory')
+	flags.DEFINE_integer('batch_size', 15, 'training batch size')
+	flags.DEFINE_integer('input_height', 256, 'input image height')
+	flags.DEFINE_integer('input_width', 256, 'input image width')
+	flags.DEFINE_integer('num_classes', 2, 'number of classes')
+	flags.DEFINE_integer('ignore_label', 254, 'label pixel value that should be ignored')
+	flags.DEFINE_boolean('random_scale', True, 'whether to perform random scaling data-augmentation')
+	flags.DEFINE_boolean('random_mirror', True, 'whether to perform random left-right flipping data-augmentation')
+
+	# log
+	flags.DEFINE_string('modeldir', 'model3', 'model directory')
+	flags.DEFINE_string('logfile', 'log.txt', 'training log filename')
+	flags.DEFINE_string('logdir', 'log', 'training log directory')
+
+	flags.FLAGS.__dict__['__parsed'] = False
+	return flags.FLAGS
+
+def main(_):
+	parser = argparse.ArgumentParser()
+	parser.add_argument('--option', dest='option', type=str, default='train',
+		help='actions: train, test, or predict')
+	args = parser.parse_args()
+
+	if args.option not in ['train', 'test', 'predict']:
+		print('invalid option: ', args.option)
+		print("Please input a option: train, test, or predict")
+	else:
+		# Set up tf session and initialize variables. 
+		# config = tf.ConfigProto()
+		# config.gpu_options.allow_growth = True
+		# sess = tf.Session(config=config)
+		sess = tf.Session()
+		# Run
+		model = Model(sess, configure())
+		getattr(model, args.option)()
+
+
+if __name__ == '__main__':
+	# Choose which gpu or cpu to use
+	os.environ['CUDA_VISIBLE_DEVICES'] = '0'
+	tf.app.run()