小编jlh*_*lhw的帖子

我正在使用TensorFlow训练卷积神经网络,将建筑物图像分为5类.

Training dataset:  
Class 1 - 3000 images
Class 2 - 3000 images
Class 3 - 3000 images
Class 4 - 3000 images
Class 5 - 3000 images

我从一个非常简单的架构开始:

Input image - 256 x 256 x 3

Convolutional layer 1 - 128 x 128 x 16 (3x3 filters, 16 filters, stride=2)
Convolutional layer 2 - 64 x 64 x 32 (3x3 filters, 32 filters, stride=2)
Convolutional layer 3 - 32 x 32 x 64 (3x3 filters, 64 filters, …

import tensorflow as tf
sess = tf.Session()

def add_to_batch(image):

    print('Adding to batch')
    image_batch = tf.train.shuffle_batch([image],batch_size=5,capacity=11,min_after_dequeue=1,num_threads=1)

    # Add to summary
    tf.image_summary('images',image_batch)

    return image_batch

def get_batch():

    # Create filename queue of images to read
    filenames = [('/media/jessica/Jessica/TensorFlow/Practice/unlabeled_data_%d.png' % i) for i in range(11)]
    filename_queue = tf.train.string_input_producer(filenames)
    reader = tf.WholeFileReader()
    key, value = reader.read(filename_queue)

    # Read and process image
    my_image = tf.image.decode_png(value)
    my_image_float = tf.cast(my_image,tf.float32)
    image_mean = tf.reduce_mean(my_image_float)
    my_noise = tf.random_normal([96,96,3],mean=image_mean)
    my_image_noisy = my_image_float + my_noise
    print('Reading images')

    return add_to_batch(my_image_noisy)

def main (): …