Merge pull request GantMan#14 from GantMan/cleanup_structure

Cleanup structure - Allows for more model types

.gitignore

@@ -2,3 +2,4 @@
 *.h5
 logs/
 .vscode/
+__pycache__

README.md

@@ -51,11 +51,24 @@ If you'd like to [say thanks for creating this, I'll take a donation for hosting
 * [Tensorflow 299x299 Image Model](https://s3.amazonaws.com/nsfwdetector/nsfw.299x299.pb)
 * _Contribute Here?  Convert the model!_
 
-## Repo Contents
+## Training Folders
 Simple description of the scripts used to create this model:
-* `train_inception_model.py` - The code used to train the Keras based Inception V3 Transfer learned model.
+* `inceptionv3_transfer/` - Folder with all the code to train the Keras based Inception v3 transfer learning model.  Includes `constants.py` for configuration, and two scripts for actual training/refinement.
 * `visuals.py` - The code to create the confusion matrix graphic
-* `self_clense.py` - The training data came down with some significant inaccuracy.  Self clense helped me use early iterations of the mode, to cross validate errors in the training data in reasonable time.   The better the model got, the better I could use it to clean the training data manually.  Most importantly, this also allowed me to clean the validation dataset, and get a real indication of generalized performance.
+* `self_clense.py` - If the training data has significant inaccuracy, `self_clense` helps cross validate errors in the training data in reasonable time.   The better the model gets, the better you can use it to clean the training data manually.
+
+_e.g._
+```bash
+cd training
+# Start with all locked transfer of Inception v3
+python inceptionv3_transfer/train_initialization.py
+
+# Continue training on model with fine-tuning
+python inceptionv3_transfer/train_fine_tune.py
+
+# Create a confusion matrix of the model
+python visuals.py
+```
 
 ## Extra Info
 There's no easy way to distribute the training data, but if you'd like to help with this model or train other models, get in touch with me and we can work together.

training/inceptionv3_transfer/callbacks.py

@@ -0,0 +1,38 @@
+from keras.callbacks import ModelCheckpoint, TensorBoard, LearningRateScheduler
+from time import time
+
+# Slow down training deeper into dataset
+def schedule(epoch):
+    if epoch < 6:
+        # Warmup model first
+        return .0000032
+    elif epoch < 12:
+        return .01
+    elif epoch < 20:
+        return .002
+    elif epoch < 40:
+        return .0004
+    elif epoch < 60:
+        return .00008
+    elif epoch < 80:
+        return .000016
+    elif epoch < 95:
+        return .0000032        
+    else:
+        return .0000009       
+
+
+def make_callbacks(weights_file):
+    # checkpoint
+    filepath = weights_file
+    checkpoint = ModelCheckpoint(
+        filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max')
+
+    # Update info
+    tensorboard = TensorBoard(log_dir="logs/{}".format(time()))
+
+    # learning rate schedule
+    lr_scheduler = LearningRateScheduler(schedule)
+
+    # all the goodies
+    return [lr_scheduler, checkpoint, tensorboard]

training/inceptionv3_transfer/constants.py

@@ -0,0 +1,12 @@
+# Config
+SIZES = {
+    'basic': 299
+}
+
+NUM_CHANNELS = 3
+NUM_CLASSES = 5
+GENERATOR_BATCH_SIZE = 32
+TOTAL_EPOCHS = 100
+STEPS_PER_EPOCH = 500
+VALIDATION_STEPS = 50
+BASE_DIR = 'D:\\nswf_model_training_data\\data'

training/inceptionv3_transfer/generators.py

@@ -0,0 +1,40 @@
+import os
+from keras.preprocessing.image import ImageDataGenerator
+import constants
+
+train_datagen = ImageDataGenerator(
+    rescale=1./255,
+    rotation_range=30,
+    width_shift_range=0.2,
+    height_shift_range=0.2,
+    shear_range=0.2,
+    zoom_range=0.2,
+    channel_shift_range=20,
+    horizontal_flip=True,
+    fill_mode='nearest'
+)
+
+# Validation data should not be modified
+validation_datagen = ImageDataGenerator(
+    rescale=1./255
+)
+
+train_dir = os.path.join(constants.BASE_DIR, 'train')
+test_dir = os.path.join(constants.BASE_DIR, 'test')
+
+def create_generators(height, width):
+    train_generator = train_datagen.flow_from_directory(
+        train_dir,
+        target_size=(height, width),
+        class_mode='categorical',
+        batch_size=constants.GENERATOR_BATCH_SIZE
+    )
+
+    validation_generator = validation_datagen.flow_from_directory(
+        test_dir,
+        target_size=(height, width),
+        class_mode='categorical',
+        batch_size=constants.GENERATOR_BATCH_SIZE
+    )
+
+    return[train_generator, validation_generator]

training/inceptionv3_transfer/train_fine_tune.py

@@ -0,0 +1,76 @@
+import os
+from keras.preprocessing.image import ImageDataGenerator
+from keras.backend import clear_session
+from keras.optimizers import SGD
+from pathlib import Path
+from keras.models import Sequential, Model, load_model
+
+# reusable stuff
+import constants
+import callbacks
+import generators
+
+# No kruft plz
+clear_session()
+
+# Config
+height = constants.SIZES['basic']
+width = height
+weights_file = "weights.best_inception" + str(height) + ".hdf5"
+
+print ('Starting from last full model run')
+model = load_model("nsfw." + str(width) + "x" + str(height) + ".h5")
+
+# Unlock a few layers deep in Inception v3
+model.trainable = False
+set_trainable = False
+for layer in model.layers:
+    if layer.name == 'conv2d_56':
+        set_trainable = True
+    if set_trainable:
+        layer.trainable = True
+    else:
+        layer.trainable = False
+
+# Let's see it
+print('Summary')
+print(model.summary())
+
+# Load checkpoint if one is found
+if os.path.exists(weights_file):
+        print ("loading ", weights_file)
+        model.load_weights(weights_file)
+
+# Get all model callbacks
+callbacks_list = callbacks.make_callbacks(weights_file)
+
+print('Compile model')
+opt = SGD(momentum=.9)
+model.compile(
+    loss='categorical_crossentropy',
+    optimizer=opt,
+    metrics=['accuracy']
+)
+
+# Get training/validation data via generators
+train_generator, validation_generator = generators.create_generators(height, width)
+
+print('Start training!')
+history = model.fit_generator(
+    train_generator,
+    callbacks=callbacks_list,
+    epochs=constants.TOTAL_EPOCHS,
+    steps_per_epoch=constants.STEPS_PER_EPOCH,
+    shuffle=True,
+    # having crazy threading issues
+    # set workers to zero if you see an error like: 
+    # `freeze_support()`
+    workers=0,
+    use_multiprocessing=True,
+    validation_data=validation_generator,
+    validation_steps=constants.VALIDATION_STEPS
+)
+
+# Save it for later
+print('Saving Model')
+model.save("nsfw." + str(width) + "x" + str(height) + ".h5")

training/inceptionv3_transfer/train_initialization.py

@@ -0,0 +1,91 @@
+import os
+from keras.preprocessing.image import ImageDataGenerator
+from keras.backend import clear_session
+from keras.optimizers import SGD
+from pathlib import Path
+from keras.applications import InceptionV3
+from keras.models import Sequential, Model, load_model
+from keras.layers import Dense, Dropout, Flatten, AveragePooling2D
+from keras import initializers, regularizers
+
+# reusable stuff
+import constants
+import callbacks
+import generators
+
+# No kruft plz
+clear_session()
+
+# Config
+height = constants.SIZES['basic']
+width = height
+weights_file = "weights.best_inception" + str(height) + ".hdf5"
+
+conv_base = InceptionV3(
+    weights='imagenet', 
+    include_top=False, 
+    input_shape=(height, width, constants.NUM_CHANNELS)
+)
+
+# First time run, no unlocking
+conv_base.trainable = False
+
+# Let's see it
+print('Summary')
+print(conv_base.summary())
+
+# Let's construct that top layer replacement
+x = conv_base.output
+x = AveragePooling2D(pool_size=(8, 8))(x)
+x - Dropout(0.4)(x)
+x = Flatten()(x)
+x = Dense(256, activation='relu', kernel_initializer=initializers.he_normal(seed=None), kernel_regularizer=regularizers.l2(.0005))(x)
+x = Dropout(0.5)(x)
+# Essential to have another layer for better accuracy
+x = Dense(128,activation='relu', kernel_initializer=initializers.he_normal(seed=None))(x)
+x = Dropout(0.25)(x)
+predictions = Dense(constants.NUM_CLASSES,  kernel_initializer="glorot_uniform", activation='softmax')(x)
+
+print('Stacking New Layers')
+model=Model(inputs = conv_base.input, outputs=predictions)
+
+# Load checkpoint if one is found
+if os.path.exists(weights_file):
+        print ("loading ", weights_file)
+        model.load_weights(weights_file)
+
+# Get all model callbacks
+callbacks_list = callbacks.make_callbacks(weights_file)
+
+print('Compile model')
+# originally adam, but research says SGD with scheduler
+# opt = Adam(lr=0.001, amsgrad=True)
+opt = SGD(momentum=.9)
+model.compile(
+    loss='categorical_crossentropy',
+    optimizer=opt,
+    metrics=['accuracy']
+)
+
+# Get training/validation data via generators
+train_generator, validation_generator = generators.create_generators(height, width)
+
+print('Start training!')
+history = model.fit_generator(
+    train_generator,
+    callbacks=callbacks_list,
+    epochs=constants.TOTAL_EPOCHS,
+    steps_per_epoch=constants.STEPS_PER_EPOCH,
+    shuffle=True,
+    # having crazy threading issues
+    # set workers to zero if you see an error like: 
+    # `freeze_support()`
+    workers=0,
+    use_multiprocessing=True,
+    validation_data=validation_generator,
+    validation_steps=constants.VALIDATION_STEPS
+)
+
+# Save it for later
+print('Saving Model')
+model.save("nsfw." + str(width) + "x" + str(height) + ".h5")

training/self_clense.py

@@ -19,8 +19,8 @@
 
 # CONFIGURE EACH RUN
 group = 'train'
-category_id = 3
-mistaken_as = 4
+category_id = 4
+mistaken_as = 2
 file_type = "jpg"