import os, sys import numpy as np import pandas as pd import matplotlib.pyplot as plt import skimage.io from skimage.transform import resize from imgaug import augmenters as iaa from tqdm import tqdm import PIL from PIL import Image import cv2 from sklearn.utils import class_weight, shuffle import warnings warnings.filterwarnings("ignore") SIZE = 299 # Load dataset info path_to_train = '../input/train/' data = pd.read_csv('../input/train.csv') train_dataset_info = [] for name, labels in zip(data['Id'], data['Target'].str.split(' ')): train_dataset_info.append({ 'path':os.path.join(path_to_train, name), 'labels':np.array([int(label) for label in labels])}) train_dataset_info = np.array(train_dataset_info) class data_generator: def create_train(dataset_info, batch_size, shape, augument=True): assert shape[2] == 3 while True: dataset_info = shuffle(dataset_info) for start in range(0, len(dataset_info), batch_size): end = min(start + batch_size, len(dataset_info)) batch_images = [] X_train_batch = dataset_info[start:end] batch_labels = np.zeros((len(X_train_batch), 28)) for i in range(len(X_train_batch)): image = data_generator.load_image( X_train_batch[i]['path'], shape) if augument: image = data_generator.augment(image) batch_images.append(image/255.) batch_labels[i][X_train_batch[i]['labels']] = 1 yield np.array(batch_images, np.float32), batch_labels def load_image(path, shape): image_red_ch = Image.open(path+'_red.png') image_yellow_ch = Image.open(path+'_yellow.png') image_green_ch = Image.open(path+'_green.png') image_blue_ch = Image.open(path+'_blue.png') image = np.stack(( np.array(image_red_ch), np.array(image_green_ch), np.array(image_blue_ch)), -1) image = cv2.resize(image, (shape[0], shape[1])) return image def augment(image): augment_img = iaa.Sequential([ iaa.OneOf([ iaa.Affine(rotate=0), iaa.Affine(rotate=90), iaa.Affine(rotate=180), iaa.Affine(rotate=270), iaa.Fliplr(0.5), iaa.Flipud(0.5), ])], random_order=True) image_aug = augment_img.augment_image(image) return image_aug from keras.preprocessing.image import ImageDataGenerator from keras.models import Sequential, load_model from keras.layers import Activation, Dropout, Flatten, Dense, GlobalMaxPooling2D, BatchNormalization, Input, Conv2D from keras.applications.inception_v3 import InceptionV3 from keras.callbacks import ModelCheckpoint from keras import metrics from keras.optimizers import Adam from keras import backend as K import keras from keras.models import Model def create_model(input_shape, n_out): input_tensor = Input(shape=input_shape) base_model = InceptionV3(include_top=False, weights='imagenet', input_shape=input_shape) bn = BatchNormalization()(input_tensor) x = base_model(bn) x = Conv2D(32, kernel_size=(1,1), activation='relu')(x) x = Flatten()(x) x = Dropout(0.5)(x) x = Dense(1024, activation='relu')(x) x = Dropout(0.5)(x) output = Dense(n_out, activation='sigmoid')(x) model = Model(input_tensor, output) return model # create callbacks list from keras.callbacks import ModelCheckpoint, LearningRateScheduler, EarlyStopping, ReduceLROnPlateau from sklearn.model_selection import train_test_split epochs = 10; batch_size = 16 checkpoint = ModelCheckpoint('../working/InceptionV3.h5', monitor='val_loss', verbose=1, save_best_only=True, mode='min', save_weights_only = True) reduceLROnPlat = ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=3, verbose=1, mode='auto', epsilon=0.0001) early = EarlyStopping(monitor="val_loss", mode="min", patience=6) callbacks_list = [checkpoint, early, reduceLROnPlat] # split data into train, valid indexes = np.arange(train_dataset_info.shape[0]) np.random.shuffle(indexes) train_indexes, valid_indexes = train_test_split(indexes, test_size=0.15, random_state=8) # create train and valid datagens train_generator = data_generator.create_train( train_dataset_info[train_indexes], batch_size, (SIZE,SIZE,3), augument=True) validation_generator = data_generator.create_train( train_dataset_info[valid_indexes], 32, (SIZE,SIZE,3), augument=False) # warm up model model = create_model( input_shape=(SIZE,SIZE,3), n_out=28) for layer in model.layers: layer.trainable = False model.layers[-1].trainable = True model.layers[-2].trainable = True model.layers[-3].trainable = True model.layers[-4].trainable = True model.layers[-5].trainable = True model.layers[-6].trainable = True model.compile( loss='binary_crossentropy', optimizer=Adam(1e-03), metrics=['acc']) # model.summary() model.fit_generator( train_generator, steps_per_epoch=np.ceil(float(len(train_indexes)) / float(batch_size)), validation_data=validation_generator, validation_steps=np.ceil(float(len(valid_indexes)) / float(batch_size)), epochs=2, verbose=1) # train all layers for layer in model.layers: layer.trainable = True model.compile(loss='binary_crossentropy', optimizer=Adam(lr=1e-4), metrics=['accuracy']) model.fit_generator( train_generator, steps_per_epoch=np.ceil(float(len(train_indexes)) / float(batch_size)), validation_data=validation_generator, validation_steps=np.ceil(float(len(valid_indexes)) / float(batch_size)), epochs=epochs, verbose=1, callbacks=callbacks_list) # Create submit submit = pd.read_csv('../input/sample_submission.csv') predicted = [] draw_predict = [] model.load_weights('../working/InceptionV3.h5') for name in tqdm(submit['Id']): path = os.path.join('../input/test/', name) image = data_generator.load_image(path, (SIZE,SIZE,3))/255. score_predict = model.predict(image[np.newaxis])[0] draw_predict.append(score_predict) label_predict = np.arange(28)[score_predict>=0.2] str_predict_label = ' '.join(str(l) for l in label_predict) predicted.append(str_predict_label) submit['Predicted'] = predicted np.save('draw_predict_InceptionV3.npy', score_predict) submit.to_csv('submit_InceptionV3.csv', index=False)