ImageNetPos.py

#!/usr/bin/env python
#-*- coding:utf-8 -*-

import tensorflow as tf 
import numpy as np 
import os 

def _activation_summary(x):
    tensor_name = x.op.name
    tf.summary.histogram(tensor_name + '/activations', x)
    tf.summary.scalar(tensor_name + '/sparsity', tf.nn.zero_fraction(x))


def ImageNetPos(_X, _weights, _biases,  dropout_prob = 1.0):
    # Reshape input picture
    _X = tf.reshape(_X, shape=[-1, 256, 256, 3])
    tf.summary.image('Image_Pos', _X, 20)
    # Layer 1
    with tf.name_scope('Image_Pos_L1') as scope:
        # Convolution Layer 1
        conv1 = tf.nn.conv2d(_X,  _weights['wc1'], strides = [1, 3, 3, 1], padding = 'VALID', name = 'conv1')
        relu1 = tf.nn.relu(tf.nn.bias_add(conv1, _biases['bc1']), name = 'relu1')
        # Max Pooling (down-sampling)
        pool1 = tf.nn.max_pool(relu1, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='VALID', name='pool1')
        _activation_summary(pool1)
    # Layer 2
    with tf.name_scope('Image_Pos_L2') as scope:
        # Convolution Layer 2
        conv2 = tf.nn.conv2d(pool1, _weights['wc2'], strides = [1, 1, 1, 1], padding='VALID', name='conv2')
        relu2 = tf.nn.relu(tf.nn.bias_add(conv2, _biases['bc2']), name='relu2')
        # Max Pooling (down-sampling)
        pool2 = tf.nn.max_pool(relu2, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='VALID', name='pool2')
        _activation_summary(pool2)
    # Layer 3
    with tf.name_scope('Image_Pos_L3') as scope:
        # Convolution Layer 3
        conv3 = tf.nn.conv2d(pool2, _weights['wc3'], [1, 1, 1, 1], padding='SAME', name='conv3')
        relu3 = tf.nn.relu(tf.nn.bias_add(conv3, _biases['bc3']), name='relu3')
        _activation_summary(relu3)
    # Layer 4
    with tf.name_scope('Image_Pos_L4') as scope:
        #Convolution Layer 4
        conv4 = tf.nn.conv2d(relu3, _weights['wc4'], [1, 1, 1, 1], padding='SAME', name='conv4')
        relu4 = tf.nn.relu(tf.nn.bias_add(conv4, _biases['bc4']), name='relu4')
        _activation_summary(relu4)

    # Layer 5
    with tf.name_scope('Image_Pos_L5') as scope:
        # Convolution Layer 5
        conv5 = tf.nn.conv2d(relu4, _weights['wc5'], [1, 1, 1, 1], padding='SAME', name='conv5')
        relu5 = tf.nn.relu(tf.nn.bias_add(conv5, _biases['bc5']), name='relu5')
        pool5 = tf.nn.max_pool(relu5, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='VALID', name='pool5')
        _activation_summary(pool5)
    # Layer 6
    with tf.name_scope('Image_Pos_L6') as scope:
        # Fully connected layer 1
        #fc6 = tf.nn.conv2d(pool5, _weights['wd1'], [1, 1, 1, 1], padding='VALID', name='fc6')
        #relu6 = tf.nn.relu(tf.nn.bias_add(fc6, _biases['bd1']), name='relu6')
        pool5_flat = tf.reshape(pool5, [-1, 8*8*256])
        dropout6 = tf.nn.dropout(pool5_flat, keep_prob=dropout_prob, name='dropout6')
        relu6 = tf.nn.relu( tf.matmul(dropout6, _weights['wd1']) + _biases['bd1']) 
        _activation_summary(dropout6)
    # Layer 7
    with tf.name_scope('Image_Pos_L7') as scope:
        # Fully connected layer 2
        #fc7 = tf.nn.conv2d(dropout6, _weights['wd2'], [1, 1, 1, 1], padding='VALID', name='fc7')
        #relu7 = tf.nn.relu(tf.nn.bias_add(fc7, _biases['bd2']), name='relu7')
        #dropout7 = tf.nn.dropout(relu7, keep_prob=dropout_prob, name='dropout7')
        relu7 = tf.nn.relu( tf.matmul(relu6, _weights['wd2']) + _biases['bd2'])
        dense2 = tf.nn.l2_normalize(relu7, dim = 1, name='fc2')
        _activation_summary(dense2)
    return dense2