baseline.py

from collections import defaultdict
from glob import glob
from random import choice, sample

import cv2
import numpy as np
import pandas as pd
from keras.applications.xception import Xception, preprocess_input
from keras.callbacks import ModelCheckpoint, ReduceLROnPlateau
from keras.layers import Input, Dense, GlobalMaxPool2D, GlobalAvgPool2D, Concatenate, Multiply, Dropout, Subtract
from keras.models import Model
from keras.optimizers import Adam
"""
from keras import backend as K
K.tensorflow_backend._get_available_gpus()

import tensorflow as tf
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
"""
train_file_path = "../../train_relationships.csv"
train_folders_path = "../../train/"
val_famillies = "F09"

all_images = glob(train_folders_path + "*/*/*.jpg")

train_images = [x for x in all_images if val_famillies not in x]
val_images = [x for x in all_images if val_famillies in x]

train_person_to_images_map = defaultdict(list)

ppl = [x.split("/")[-3] + "/" + x.split("/")[-2] for x in all_images]

for x in train_images:
    train_person_to_images_map[x.split("/")[-3] + "/" + x.split("/")[-2]].append(x)

val_person_to_images_map = defaultdict(list)

for x in val_images:
    val_person_to_images_map[x.split("/")[-3] + "/" + x.split("/")[-2]].append(x)

relationships = pd.read_csv(train_file_path)
relationships = list(zip(relationships.p1.values, relationships.p2.values))
relationships = [x for x in relationships if x[0] in ppl and x[1] in ppl]

train = [x for x in relationships if val_famillies not in x[0]]
val = [x for x in relationships if val_famillies in x[0]]


def read_img(path):
    img = cv2.imread(path)
    return preprocess_input(img)


def gen(list_tuples, person_to_images_map, batch_size=16):
    ppl = list(person_to_images_map.keys())
    while True:
        batch_tuples = sample(list_tuples, batch_size // 2)
        labels = [1] * len(batch_tuples)
        while len(batch_tuples) < batch_size:
            p1 = choice(ppl)
            p2 = choice(ppl)

            if p1 != p2 and (p1, p2) not in list_tuples and (p2, p1) not in list_tuples:
                batch_tuples.append((p1, p2))
                labels.append(0)

        for x in batch_tuples:
            if not len(person_to_images_map[x[0]]):
                print(x[0])

        X1 = [choice(person_to_images_map[x[0]]) for x in batch_tuples]
        X1 = np.array([read_img(x) for x in X1])

        X2 = [choice(person_to_images_map[x[1]]) for x in batch_tuples]
        X2 = np.array([read_img(x) for x in X2])

        yield [X1, X2], labels


def baseline_model():
    input_1 = Input(shape=(224, 224, 3))
    input_2 = Input(shape=(224, 224, 3))

    base_model = Xception(weights='imagenet', include_top=False)
    #base_model.summary()
    for x in base_model.layers[:-3]:
        x.trainable = True 

    x1 = base_model(input_1)
    x2 = base_model(input_2)
	
    print(x1.shape)
    print(x2.shape)

    # x1_ = Reshape(target_shape=(7*7, 2048))(x1)
    # x2_ = Reshape(target_shape=(7*7, 2048))(x2)
    #
    # x_dot = Dot(axes=[2, 2], normalize=True)([x1_, x2_])
    # x_dot = Flatten()(x_dot)

    x1 = Concatenate(axis=-1)([GlobalMaxPool2D()(x1), GlobalAvgPool2D()(x1)])
    x2 = Concatenate(axis=-1)([GlobalMaxPool2D()(x2), GlobalAvgPool2D()(x2)])

    x3 = Subtract()([x1, x2])
    x3 = Multiply()([x3, x3])

    x = Multiply()([x1, x2])

    x = Concatenate(axis=-1)([x, x3])

    x = Dense(100, activation="relu")(x)
    x = Dropout(0.01)(x)
    out = Dense(1, activation="sigmoid")(x)

    model = Model([input_1, input_2], out)

    model.compile(loss="binary_crossentropy", metrics=['acc'], optimizer=Adam(0.00001))

    model.summary()

    return model


file_path = "baseline.h5"

checkpoint = ModelCheckpoint(file_path, monitor='val_acc', verbose=1, save_best_only=True, mode='max')

reduce_on_plateau = ReduceLROnPlateau(monitor="val_acc", mode="max", factor=0.1, patience=20, verbose=1)

callbacks_list = [checkpoint, reduce_on_plateau]

model = baseline_model()
# model.load_weights(file_path)
model.fit_generator(gen(train, train_person_to_images_map, batch_size=16), use_multiprocessing=True,
                    validation_data=gen(val, val_person_to_images_map, batch_size=16), epochs=100, verbose=2,
                    workers=4, callbacks=callbacks_list, steps_per_epoch=200, validation_steps=100)

test_path = "../../test/"


def chunker(seq, size=32):
    return (seq[pos:pos + size] for pos in range(0, len(seq), size))


from tqdm import tqdm

submission = pd.read_csv('../../sample_submission.csv')

predictions = []

for batch in tqdm(chunker(submission.img_pair.values)):
    X1 = [x.split("-")[0] for x in batch]
    X1 = np.array([read_img(test_path + x) for x in X1])

    X2 = [x.split("-")[1] for x in batch]
    X2 = np.array([read_img(test_path + x) for x in X2])

    pred = model.predict([X1, X2]).ravel().tolist()
    predictions += pred

submission['is_related'] = predictions

submission.to_csv("baseline.csv", index=False)