Propose a training methodology for GANs where start with low-resolution images, and then progressively increase the resolution by adding layers to the networks.
Increasing variation using minibatch standard deviation
A new normalization in G and D
Propose sliced Wasserstein distance (SWD) to estimate the statistical similarity.
Most of current object detection methods only use local image patch independently. However, the critical correlation cues among different proposals (e.g., relative spatial layouts or semantic correlations) are often ignored.
In this paper, in order to exploit global interdependency among objects, they propose a Tree-RL approach that learns to localize multiple objects sequentially based on both the current observation and historical search paths.
The basic GAN model was proposed by Ian Goodfellow in 2014.
This model include 2 Parts: generative model and discriminate model:
The generative model intend to produce the sample close to real sample.
The discriminator intend to improve the ability to judge whether the sample is real.
编程语言是用来定义计算机程序的形式语言。它是一种被标准化的交流技巧，用来向计算机发出指令。一种计算机语言让程序员能够准确地定义计算机所需要使用的数据，并精确地定义在不同情况下所应当采取的行动。
编程语言是具有不同层次的，有机器语言，汇编语言，高级语言。要理解编程语言的层次，我们要从什么是编程说起。
主流的深度学习算法分为两种: 生成模型和判别模型。其中：
对联合分布 P(X,Y)进行建模。 X为输入，Y为输出或者隐含层。
直观来讲，我们认为 输入的数据X和输出的标签或者回归值Y合并的变量构成一个分布P(X,Y)。而我们所有的训练样本 (X,Y)均是由这个分布生成。对于生成模型的目的，就是学习到这样一个分布，也可以说是(X,Y)的联合分布。生成模型可以得到模型中任何一个变量的值。
直接对 Y=f(X)或者 P(Y|X)建模。
而对于判别模型，我们认为输入数据 X 和输出值 Y构成函数关系，而判别模型的任务就是学习直接学习到这个函数关系，从而直接由输入X估计输出值Y。
判别模型只能再给定输入的情况下得到输出的值。
This is common for all deep learning work. And it depends on the data. In python, many people like to use pickle to store data.
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/",one_hot = True)
In general, data include inputs and labels.