Joseph cleanup

.editorconfig

@@ -0,0 +1,21 @@
+# EditorConfig helps developers define and maintain consistent
+# coding styles between different editors and IDEs
+# editorconfig.org
+
+root = true
+
+
+[*]
+
+# Change these settings to your own preference
+indent_style = space
+indent_size = 4
+
+# We recommend you to keep these unchanged
+end_of_line = lf
+charset = utf-8
+trim_trailing_whitespace = true
+insert_final_newline = true
+
+[*.md]
+trim_trailing_whitespace = false

.pylintrc

@@ -0,0 +1,18 @@
+#My pylintrc for use with atom.io's linter-pylint
+[MESSAGES CONTROL]
+disable=W0311,W1201,W0702,W0603,W0611,W0621,W0703,W0212,E1101,E0611,C0111,C0103,R0902,R0914,I0011
+
+# checks for :
+# * unauthorized constructions
+# * strict indentation
+# * line length
+# * use of <> instead of !=
+#
+[FORMAT]
+# Maximum number of characters on a single line.
+max-line-length=128
+# Maximum number of lines in a module
+max-module-lines=1000
+# String used as indentation unit. This is usually "    " (4 spaces) or "\t" (1
+# tab).  In repo it is 2 spaces.
+indent-string='    '

bootstrap.py

@@ -1,15 +1,18 @@
 #!/usr/bin/env python
+
+from __future__ import print_function
+
 import os
 import glob
+import sys
 import tarfile
+from shutil import copyfile, rmtree
 import numpy as np
 from scipy.io import loadmat
-from shutil import copyfile, rmtree
-import sys
 import config
 
 if sys.version_info[0] >= 3:
-    from urllib.request import urlretrieve
+    from urllib.request import urlretrieve # #pylint: disable=W,E
 else:
     # Not Python 3 - today, it is most likely to be Python 2
     # But note that this might need an update when Python 4
@@ -31,7 +34,7 @@ def download_file(url, dest=None):
 
 flowers_archive_path = os.path.join(data_path, '102flowers.tgz')
 if not os.path.isfile(flowers_archive_path):
-    print ('Downloading images...')
+    print('Downloading images...')
     download_file('http://www.robots.ox.ac.uk/~vgg/data/flowers/102/102flowers.tgz')
 tarfile.open(flowers_archive_path).extractall(path=data_path)
 

config.py

@@ -8,7 +8,7 @@
 data_dir = join_path(abspath, 'data/sorted')
 trained_dir = join_path(abspath, 'trained')
 
-train_dir, validation_dir = None, None
+train_dir, validation_dir, test_dir = None, None, None
 
 MODEL_VGG16 = 'vgg16'
 MODEL_INCEPTION_V3 = 'inception_v3'
@@ -37,12 +37,13 @@
 
 nb_train_samples = 0
 nb_validation_samples = 0
-
+nb_test_samples = 0
 
 def set_paths():
-    global train_dir, validation_dir
+    global train_dir, validation_dir, test_dir
     train_dir = join_path(data_dir, 'train/')
     validation_dir = join_path(data_dir, 'valid/')
+    test_dir = join_path(data_dir, 'test/')
 
 
 set_paths()

download_pretrained_resnet50.sh

@@ -0,0 +1,13 @@
+#!/bin/sh
+
+DIR=$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )
+
+cd $DIR
+mkdir trained
+cd trained
+
+HOST='https://drop.jackhftang.com/storage'
+curl -LO "$HOST/model-resnet50.h5"
+curl -LO "$HOST/fine-tuned-resnet50-weights.h5"
+curl -LO "$HOST/classes-resnet50"
+

models/base_model.py

@@ -1,3 +1,5 @@
+from __future__ import print_function
+
 from keras.callbacks import EarlyStopping, ModelCheckpoint
 from keras.preprocessing import image
 from keras.applications.imagenet_utils import preprocess_input
@@ -43,7 +45,8 @@ def _fine_tuning(self):
                 train_data,
                 steps_per_epoch=config.nb_train_samples / float(self.batch_size),
                 epochs=self.nb_epoch,
-                validation_data=self.get_validation_datagen(),
+                validation_data=self.get_validation_datagen(rotation_range=30., shear_range=0.2,
+                                                            zoom_range=0.2, horizontal_flip=True),
                 validation_steps=config.nb_validation_samples / float(self.batch_size),
                 callbacks=callbacks,
                 class_weight=self.class_weight)
@@ -52,7 +55,8 @@ def _fine_tuning(self):
                 train_data,
                 samples_per_epoch=config.nb_train_samples,
                 nb_epoch=self.nb_epoch,
-                validation_data=self.get_validation_datagen(),
+                validation_data=self.get_validation_datagen(rotation_range=30., shear_range=0.2,
+                                                            zoom_range=0.2, horizontal_flip=True),
                 nb_val_samples=config.nb_validation_samples,
                 callbacks=callbacks,
                 class_weight=self.class_weight)

predict.py

@@ -1,8 +1,10 @@
+from __future__ import print_function
+
 import time
 import argparse
 import os
-import numpy as np
 import glob
+import numpy as np
 from sklearn.metrics import accuracy_score, confusion_matrix
 from sklearn.externals import joblib
 

requirements.txt

@@ -8,4 +8,5 @@ scikit_learn
 theano==0.9.0
 h5py
 Pillow
-tensorflow-gpu
+tensorflow-gpu
+Augmentor

server.py

@@ -1,12 +1,14 @@
+from __future__ import print_function
+
+import os
+import argparse
 import socket
+import traceback
 from threading import Thread
 import numpy as np
-import os
-import argparse
 import config
 import util
 from sklearn.externals import joblib
-import traceback
 
 util.set_img_format()
 
@@ -60,7 +62,7 @@ def handle(clientsocket):
                     predicted_relativity = novelty_detection_clf.predict(acts)[0]
                     nd_class = novelty_detection_clf.__classes[predicted_relativity]
                 except Exception as e:
-                    print(e.message)
+                    print(e)
                     nd_class = 'related'
 
                 top10_json = "["

test.py

@@ -0,0 +1,84 @@
+from __future__ import print_function
+
+import os
+import argparse
+import traceback
+import numpy as np
+
+import util
+import config
+import keras
+from keras.preprocessing.image import ImageDataGenerator
+
+np.random.seed(1337)  # for reproducibility
+batch_size=16
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--data_dir', help='Path to data dir')
+    parser.add_argument('--trained_file', default='./trained/model-resnet50.h5', help='Path to trained file')
+    parser.add_argument('--model', type=str, default='resnet50', help='Base model architecture',
+                        choices=[config.MODEL_RESNET50, config.MODEL_RESNET152, config.MODEL_INCEPTION_V3,
+                                 config.MODEL_VGG16])
+    args = parser.parse_args()
+    config.model = args.model
+    config.trained_file = args.trained_file
+    return args
+
+
+def init():
+    util.set_img_format()
+    util.set_classes_from_train_dir()
+    util.set_samples_info()
+    if not os.path.exists(config.trained_file):
+        raise Exception('trained_file not exists')
+
+def test():
+    img_size = (224, 224)
+
+    print("Creating model...")
+
+    model = keras.models.load_model(config.trained_file)
+
+    print("Model is created")
+
+    idg = ImageDataGenerator()
+    idg.mean = np.array([103.939, 116.779, 123.68], dtype=np.float32).reshape((3, 1, 1))
+    test_generator = idg.flow_from_directory(config.test_dir,
+        batch_size=batch_size,
+        target_size=img_size,
+        classes=config.classes)
+
+    # -- Evaluate generator -- #
+    result = model.evaluate_generator(
+        generator=test_generator,
+        steps=config.nb_test_samples)
+
+    print("Model [loss, accuracy]: {0}".format(result))
+
+    # -- Predict generator -- #
+    predict = model.predict_generator(
+        generator=test_generator,
+        steps=config.nb_test_samples)
+
+    print("model predictions: {0}".format(predict))
+    print('Testing is finished!')
+
+
+if __name__ == '__main__':
+    try:
+        args = parse_args()
+        if args.data_dir:
+            config.data_dir = args.data_dir
+            config.set_paths()
+        if args.model:
+            config.model = args.model
+
+        init()
+        test()
+
+    except Exception as e:
+        print(e)
+        traceback.print_exc()
+    finally:
+        util.unlock()

train.py

@@ -1,24 +1,26 @@
-import numpy as np
+from __future__ import print_function
+
+import os
 import argparse
 import traceback
-import os
-
-np.random.seed(1337)  # for reproducibility
+import numpy as np
 
 import util
 import config
 
+np.random.seed(1337)  # for reproducibility
 
 def parse_args():
     parser = argparse.ArgumentParser()
     parser.add_argument('--data_dir', help='Path to data dir')
-    parser.add_argument('--model', type=str, required=True, help='Base model architecture', choices=[
+    parser.add_argument('--model', type=str, help='Base model architecture', choices=[
         config.MODEL_RESNET50,
         config.MODEL_RESNET152,
         config.MODEL_INCEPTION_V3,
         config.MODEL_VGG16])
     parser.add_argument('--nb_epoch', type=int, default=1000)
-    parser.add_argument('--freeze_layers_number', type=int, help='will freeze the first N layers and unfreeze the rest')
+    parser.add_argument('--freeze_layers_number',
+                        type=int, help='will freeze the first N layers and unfreeze the rest')
     return parser.parse_args()
 
 
@@ -52,6 +54,7 @@ def train(nb_epoch, freeze_layers_number):
 
         init()
         train(args.nb_epoch, args.freeze_layers_number)
+
     except Exception as e:
         print(e)
         traceback.print_exc()

train_novelty_detection.py

@@ -1,3 +1,6 @@
+from __future__ import print_function
+
+import argparse
 from sklearn.model_selection import GridSearchCV
 from sklearn.preprocessing import LabelEncoder
 from sklearn.linear_model import LogisticRegression
@@ -7,7 +10,6 @@
 from keras.models import Sequential
 from keras.layers import Dense, Dropout
 from keras.callbacks import EarlyStopping, ModelCheckpoint
-import argparse
 import pandas as pd
 import numpy as np
 import config

util.py

@@ -1,22 +1,26 @@
+from __future__ import print_function
+
+import os
+import glob
+import math
+import itertools
+
+import importlib
 import matplotlib
 
 matplotlib.use('Agg')  # fixes issue if no GUI provided
+
 import matplotlib.pyplot as plt
 
 import seaborn as sns
-
-sns.set(style='white')
-
 import numpy as np
-import os
-import glob
 import pandas as pd
-import importlib
 import keras
 from keras import backend as K
+from keras.preprocessing.image import DirectoryIterator
 import config
-import math
-import itertools
+
+sns.set(style='white')
 
 
 def save_history(history, prefix):
@@ -103,16 +107,17 @@ def set_samples_info():
     """Walks through the train and valid directories
     and returns number of images"""
     white_list_formats = {'png', 'jpg', 'jpeg', 'bmp'}
-    dirs_info = {config.train_dir: 0, config.validation_dir: 0}
+    dirs_info = {config.train_dir: 0, config.validation_dir: 0, config.test_dir: 0}
     for d in dirs_info:
         iglob_iter = glob.iglob(d + '**/*.*')
         for i in iglob_iter:
-            filename, file_extension = os.path.splitext(i)
+            _, file_extension = os.path.splitext(i)
             if file_extension[1:] in white_list_formats:
                 dirs_info[d] += 1
 
     config.nb_train_samples = dirs_info[config.train_dir]
     config.nb_validation_samples = dirs_info[config.validation_dir]
+    config.nb_test_samples = dirs_info[config.test_dir]
 
 
 def get_class_weight(d):
@@ -150,8 +155,6 @@ def set_classes_from_train_dir():
 def override_keras_directory_iterator_next():
     """Overrides .next method of DirectoryIterator in Keras
       to reorder color channels for images from RGB to BGR"""
-    from keras.preprocessing.image import DirectoryIterator
-
     original_next = DirectoryIterator.next
 
     # do not allow to override one more time
@@ -192,8 +195,8 @@ def save_activations(model, inputs, files, layer, batch_number):
     all_activations = []
     ids = []
     af = get_activation_function(model, layer)
-    for i in range(len(inputs)):
-        acts = get_activations(af, [inputs[i]])
+    for i, inp in enumerate(inputs):
+        acts = get_activations(af, [inp])
         all_activations.append(acts)
         ids.append(files[i].split('/')[-2])