File and folder restructuring for more organized code

.gitignore

@@ -1,4 +1,6 @@
 *.pyc
+*.graph
+models/*.graph
 models/*.pyc
 models/*.ckpt*
 models/*.pb

models/frontend_builder.py → builders/frontend_builder.py

@@ -1,8 +1,8 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
-import resnet_v2
-import mobilenet_v2
-import inception_v4
+from frontends import resnet_v2
+from frontends import mobilenet_v2
+from frontends import inception_v4
 import os 
 
 

model_builder.py → builders/model_builder.py

@@ -3,28 +3,29 @@
 import subprocess
 
 sys.path.append("models")
-from FC_DenseNet_Tiramisu import build_fc_densenet
-from Encoder_Decoder import build_encoder_decoder
-from RefineNet import build_refinenet
-from FRRN import build_frrn
-from MobileUNet import build_mobile_unet
-from PSPNet import build_pspnet
-from GCN import build_gcn
-from DeepLabV3 import build_deeplabv3
-from DeepLabV3_plus import build_deeplabv3_plus
-from AdapNet import build_adaptnet
+from models.FC_DenseNet_Tiramisu import build_fc_densenet
+from models.Encoder_Decoder import build_encoder_decoder
+from models.RefineNet import build_refinenet
+from models.FRRN import build_frrn
+from models.MobileUNet import build_mobile_unet
+from models.PSPNet import build_pspnet
+from models.GCN import build_gcn
+from models.DeepLabV3 import build_deeplabv3
+from models.DeepLabV3_plus import build_deeplabv3_plus
+from models.AdapNet import build_adaptnet
+from models.custom_model import build_custom
 
 SUPPORTED_MODELS = ["FC-DenseNet56", "FC-DenseNet67", "FC-DenseNet103", "Encoder-Decoder", "Encoder-Decoder-Skip", "RefineNet",
     "FRRN-A", "FRRN-B", "MobileUNet", "MobileUNet-Skip", "PSPNet", "GCN", "DeepLabV3", "DeepLabV3_plus", "AdapNet", "custom"]
 
 SUPPORTED_FRONTENDS = ["ResNet50", "ResNet101", "ResNet152", "MobileNetV2", "InceptionV4"]
 
 def download_checkpoints(model_name):
-    subprocess.check_output(["python", "get_pretrained_checkpoints.py", "--model=" + model_name])
+    subprocess.check_output(["python", "utils/get_pretrained_checkpoints.py", "--model=" + model_name])
 
 
 
-def build_model(model_name, net_input, num_classes, frontend="ResNet101", is_training=True):
+def build_model(model_name, net_input, num_classes, crop_width, crop_height, frontend="ResNet101", is_training=True):
 	# Get the selected model. 
 	# Some of them require pre-trained ResNet
 
@@ -63,7 +64,7 @@ def build_model(model_name, net_input, num_classes, frontend="ResNet101", is_tra
 	elif model_name == "PSPNet":
 	    # Image size is required for PSPNet
 	    # PSPNet requires pre-trained ResNet weights
-	    network, init_fn = build_pspnet(net_input, label_size=[args.crop_height, args.crop_width], preset_model = model_name, frontend=frontend, num_classes=num_classes, is_training=is_training)
+	    network, init_fn = build_pspnet(net_input, label_size=[crop_height, crop_width], preset_model = model_name, frontend=frontend, num_classes=num_classes, is_training=is_training)
 	elif model_name == "GCN":
 	    # GCN requires pre-trained ResNet weights
 	    network, init_fn = build_gcn(net_input, preset_model = model_name, frontend=frontend, num_classes=num_classes, is_training=is_training)

models/inception_v4.py → frontends/inception_v4.py

@@ -26,7 +26,7 @@
 
 import tensorflow as tf
 
-import inception_utils
+from frontends import inception_utils
 
 slim = tf.contrib.slim
 

models/mobilenet_v2.py → frontends/mobilenet_v2.py

@@ -29,8 +29,8 @@
 
 import tensorflow as tf
 
-import conv_blocks as ops
-import mobilenet_base as lib
+from frontends import conv_blocks as ops
+from frontends import mobilenet_base as lib
 
 slim = tf.contrib.slim
 op = lib.op

models/resnet_v1.py → frontends/resnet_v1.py

@@ -1,6 +1,6 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
-import resnet_utils
+from frontends import resnet_utils
 
 resnet_arg_scope = resnet_utils.resnet_arg_scope
 

models/resnet_v2.py → frontends/resnet_v2.py

@@ -52,7 +52,7 @@
 
 import tensorflow as tf
 
-import resnet_utils
+from frontends import resnet_utils
 
 slim = tf.contrib.slim
 resnet_arg_scope = resnet_utils.resnet_arg_scope

models/AdapNet.py

@@ -3,7 +3,7 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
 import numpy as np
-import resnet_v2
+from frontends import resnet_v2
 import os, sys
 
 

models/DeepLabV3.py

@@ -3,7 +3,7 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
 import numpy as np
-import frontend_builder
+from builders import frontend_builder
 import os, sys
 
 def Upsampling(inputs,feature_map_shape):

models/DeepLabV3_plus.py

@@ -2,9 +2,8 @@
 
 import tensorflow as tf
 from tensorflow.contrib import slim
-import frontend_builder
+from builders import frontend_builder
 import numpy as np
-import resnet_v2
 import os, sys
 
 def Upsampling(inputs,feature_map_shape):

models/FRRN.py

@@ -1,6 +1,5 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
-import resnet_v1
 
 def Upsampling(inputs,scale):
     return tf.image.resize_nearest_neighbor(inputs, size=[tf.shape(inputs)[1]*scale,  tf.shape(inputs)[2]*scale])

models/GCN.py

@@ -1,6 +1,6 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
-import frontend_builder
+from builders import frontend_builder
 import os, sys
 
 def Upsampling(inputs,scale):

models/ICNet.py

@@ -1,7 +1,7 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
 import numpy as np
-import frontend_builder
+from frontends import frontend_builder
 import os, sys
 
 def Upsampling_by_shape(inputs, feature_map_shape):

models/PSPNet.py

@@ -1,7 +1,7 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
 import numpy as np
-import frontend_builder
+from builders import frontend_builder
 import os, sys
 
 def Upsampling(inputs,feature_map_shape):

models/RefineNet.py

@@ -1,6 +1,6 @@
 import tensorflow as tf
 from tensorflow.contrib import slim
-import frontend_builder
+from builders import frontend_builder
 import os, sys
 
 def Upsampling(inputs,scale):
@@ -142,7 +142,7 @@ def RefineBlock(high_inputs=None,low_inputs=None):
 
 
 
-def build_refinenet(inputs, num_classes, preset_model='RefineNet', frontend="ResNet101", weight_decay=1e-5, upscaling_method="bilinear", pretrained_dir="models"):
+def build_refinenet(inputs, num_classes, preset_model='RefineNet', frontend="ResNet101", weight_decay=1e-5, upscaling_method="bilinear", pretrained_dir="models", is_training=True):
     """
     Builds the RefineNet model. 
 

models/custom_model.py

@@ -3,7 +3,7 @@
 import tensorflow as tf
 import tensorflow.contrib.slim as slim
 import numpy as np
-import frontend_builder
+from builders import frontend_builder
 
 def conv_block(inputs, n_filters, filter_size=[3, 3], dropout_p=0.0):
 	"""

predict.py

@@ -3,9 +3,8 @@
 import argparse
 import numpy as np
 
-import utils
-import helpers
-import model_builder
+from utils import utils, helpers
+from builders import model_builder
 
 parser = argparse.ArgumentParser()
 parser.add_argument('--image', type=str, default=None, required=True, help='The image you want to predict on. ')

test.py

@@ -3,9 +3,8 @@
 import argparse
 import numpy as np
 
-import utils
-import helpers
-import model_builder
+from utils import utils, helpers
+from builders import model_builder
 
 parser = argparse.ArgumentParser()
 parser.add_argument('--checkpoint_path', type=str, default=None, required=True, help='The path to the latest checkpoint weights for your model.')

train.py

@@ -10,14 +10,11 @@
 import subprocess
 
 
-import helpers 
-import utils 
+from utils import utils, helpers
+from builders import model_builder
 
 import matplotlib.pyplot as plt
 
-sys.path.append("models")
-import model_builder
-
 def str2bool(v):
     if v.lower() in ('yes', 'true', 't', 'y', '1'):
         return True
@@ -90,7 +87,7 @@ def data_augmentation(input_image, output_image):
 net_input = tf.placeholder(tf.float32,shape=[None,None,None,3])
 net_output = tf.placeholder(tf.float32,shape=[None,None,None,num_classes]) 
 
-network, init_fn = model_builder.build_model(args.model, frontend=args.frontend, net_input=net_input, num_classes=num_classes, is_training=True)
+network, init_fn = model_builder.build_model(model_name=args.model, frontend=args.frontend, net_input=net_input, num_classes=num_classes, crop_width=args.crop_width, crop_height=args.crop_height, is_training=True)
 
 loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=network, labels=net_output))
 

utils.py → utils/utils.py

@@ -11,7 +11,7 @@
     recall_score, confusion_matrix, classification_report, \
     accuracy_score, f1_score
 
-import helpers
+from utils import helpers
 
 def prepare_data(dataset_dir):
     train_input_names=[]