input_data.py

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import os
import random
import tensorflow as tf

IMAGE_SIZE = 224
NUM_CLASSES = 101

NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 100
NUM_EXAMPLES_PER_EPOCH_FOR_EVAL = 20


def get_frame_path(videoname):
    image_path = ''
    for root, dirs, filenames in os.walk(videoname):
        index = random.randint(0, len(filenames) - 1)
        image_path = os.path.join(videoname, filenames[index])
    # img_raw =tf.gfile.FastGFile(image_name,'rb').read()
    # img = tf.image.decode_jpeg(img_raw).eval()
    return image_path


def read_data(input_queue):
    label = input_queue[1]
    file_contents = tf.read_file(input_queue[0])
    example = tf.image.decode_jpeg(file_contents, channels=3)

    return example, label


def _generate_image_and_label_batch(image, label, min_queue_examples, batch_size, shuffle):
    num_preprpocess_threads = 4
    if shuffle:
        images, label_batch = tf.train.shuffle_batch(
            [image, label],
            batch_size=batch_size,
            num_threads=num_preprpocess_threads,
            capacity=min_queue_examples + 3 * batch_size,
            min_after_dequeue=min_queue_examples)
    else:
        images, label_batch = tf.train.batch(
            [image, label],
            batch_size=batch_size,
            num_threads=num_preprpocess_threads,
            capacity=min_queue_examples + 3 * batch_size)
    return images, tf.reshape(label_batch, [batch_size])


def distorted_inputs(ucf_path, batch_size):
    image_list = []
    label_list = []
    actions = os.listdir(ucf_path)
    # print(actions)
    for i in range(len(actions)):
        videos = os.listdir(os.path.join(ucf_path, actions[i]))
        # print(videos)
        for video in videos:
            image_path = get_frame_path(os.path.join(ucf_path, actions[i], video))
            # print(image_path)
            image_list.append(image_path)
            label_list.append(i)

    labels = tf.convert_to_tensor(label_list, dtype=tf.int32)
    images = tf.convert_to_tensor(image_list, dtype=tf.string)

    input_queue = tf.train.slice_input_producer([images, labels], shuffle=False)

    with tf.name_scope('data_augmentation'):
        height = IMAGE_SIZE
        width = IMAGE_SIZE

        image, label = read_data(input_queue)
        reshaped_image = tf.cast(image, tf.float32)

        distorted_image = tf.random_crop(reshaped_image, [height, width, 3])
        distorted_image = tf.image.random_brightness(distorted_image, max_delta=63)
        distorted_image = tf.image.random_flip_left_right(distorted_image, 1)

        min_fraction_of_examples_in_queue = 0.4
        min_queue_examples = int(NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN * min_fraction_of_examples_in_queue)

        print('Filling queue with %d images before starting to train.' % min_queue_examples)
        return _generate_image_and_label_batch(distorted_image, label, min_queue_examples, batch_size, shuffle=True)

# if __name__ == '__main__':
#     with tf.Graph().as_default():
#         result = distorted_inputs(FILE_PATH, 64)
#
#         with tf.Session() as sess:
#             coord = tf.train.Coordinator()
#             threads = tf.train.start_queue_runners(coord=coord)
#             for i in range(10):
#                 example = sess.run(result)
#                 print('hello')
#             coord.request_stop()
#             coord.join(threads)