tensorflow 学习笔记： recurrent models of visual attention

tf.truncated_normal：

tf.truncated_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None)

The generated values follow a normal distribution with specified mean andstandard deviation, except thatvalues whose magnitude is more than 2 standarddeviations from the mean are dropped and re-picked.

tf.Variable():

需要初始化值和shape。后期可以通过assign改变value和shape。validate_shape=False




tf.pack():

将N个tensor按照某个维度拼接起来。可以理解为先按照行进行拼接，再进行维度的转换


class中__call__函数把类当成一个对象

tf.image.extract_glimpse(input, size, offsets, centered=None, normalized=None, uniform_noise=None, name=None)