In [1]:
GPU_NUMBER = 3
root = '/data/vision/torralba/health-habits/other/enes/'
%matplotlib inline
import matplotlib.pyplot as plt
import os
import sys
import random
import json
import math
sys.path.append( root + 'Utils/')
import pandas as pd
import numpy as np
import tensorflow as tf
from PIL import Image
from IPython.display import display
from pprint import pprint
from notebook_utils import *
from skimage import color, io
os.environ['CUDA_VISIBLE_DEVICES'] = str(GPU_NUMBER);
In [2]:
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
def conv2d(x, W, s):
return tf.nn.conv2d(x, W, strides=[1, s, s, 1], padding='SAME')
In [3]:
def setup_tensorflow_graph(BATCH_SIZE):
image_ = tf.placeholder( tf.float32, shape = [BATCH_SIZE, 256, 256, 1] )
output_ = tf.placeholder( tf.float32, shape = [BATCH_SIZE, 64, 64, 313])
W1_1 = weight_variable([3,3,1,64])
b1_1 = bias_variable([64])
conv1_1 = conv2d( image_, W1_1, 1 ) + b1_1
conv1_1 = tf.nn.relu(conv1_1)
W1_2 = weight_variable([3,3,64,64])
b1_2 = bias_variable([64])
conv1_2 = conv2d( conv1_1, W1_2, 2 ) + b1_2
conv1_2 = tf.nn.relu(conv1_2)
conv1_2 = tf.contrib.layers.batch_norm(conv1_2)
W2_1 = weight_variable([3,3,64,128])
b2_1 = bias_variable([128])
conv2_1 = conv2d( conv1_2, W2_1, 1 ) + b2_1
conv2_1 = tf.nn.relu(conv2_1)
W2_2 = weight_variable([3,3,128,128])
b2_2 = bias_variable([128])
conv2_2 = conv2d( conv2_1, W2_2, 2 ) + b2_2
conv2_2 = tf.nn.relu(conv2_2)
conv2_2 = tf.contrib.layers.batch_norm(conv2_2)
W3_1 = weight_variable([3,3,128,256])
b3_1 = bias_variable([256])
conv3_1 = conv2d( conv2_2, W3_1, 1 ) + b3_1
conv3_1 = tf.nn.relu(conv3_1)
W3_2 = weight_variable([3,3,256,256])
b3_2 = bias_variable([256])
conv3_2 = conv2d( conv3_1, W3_2, 1 ) + b3_2
conv3_2 = tf.nn.relu(conv3_2)
W3_3 = weight_variable([3,3,256,256])
b3_3 = bias_variable([256])
conv3_3 = conv2d( conv3_2, W3_3, 2 ) + b3_3
conv3_3 = tf.nn.relu(conv3_3)
conv3_3 = tf.contrib.layers.batch_norm(conv3_3)
W4_1 = weight_variable([3,3,256,512])
b4_1 = bias_variable([512])
conv4_1 = conv2d( conv3_3, W4_1, 1 ) + b4_1
conv4_1 = tf.nn.relu(conv4_1)
W4_2 = weight_variable([3,3,512,512])
b4_2 = bias_variable([512])
conv4_2 = conv2d( conv4_1, W4_2, 1 ) + b4_2
conv4_2 = tf.nn.relu(conv4_2)
W4_3 = weight_variable([3,3,512,512])
b4_3 = bias_variable([512])
conv4_3 = conv2d( conv4_2, W4_3, 1 ) + b4_3
conv4_3 = tf.nn.relu(conv4_3)
conv4_3 = tf.contrib.layers.batch_norm(conv4_3)
W5_1 = weight_variable([3,3,512,512])
b5_1 = bias_variable([512])
conv5_1 = tf.nn.atrous_conv2d( conv4_3, W5_1, 2, padding = 'SAME') + b5_1
conv5_1 = tf.nn.relu(conv5_1)
W5_2 = weight_variable([3,3,512,512])
b5_2 = bias_variable([512])
conv5_2 = tf.nn.atrous_conv2d( conv5_1, W5_2, 2, padding = 'SAME') + b5_2
conv5_2 = tf.nn.relu(conv5_2)
W5_3 = weight_variable([3,3,512,512])
b5_3 = bias_variable([512])
conv5_3 = tf.nn.atrous_conv2d( conv5_2, W5_3, 2, padding = 'SAME') + b5_3
conv5_3 = tf.nn.relu(conv5_3)
conv5_3 = tf.contrib.layers.batch_norm(conv5_3)
W6_1 = weight_variable([3,3,512,512])
b6_1 = bias_variable([512])
conv6_1 = tf.nn.atrous_conv2d( conv5_3, W6_1, 2, padding = 'SAME') + b6_1
conv6_1 = tf.nn.relu(conv6_1)
W6_2 = weight_variable([3,3,512,512])
b6_2 = bias_variable([512])
conv6_2 = tf.nn.atrous_conv2d( conv6_1, W6_2, 2, padding = 'SAME') + b6_2
conv6_2 = tf.nn.relu(conv6_2)
W6_3 = weight_variable([3,3,512,512])
b6_3 = bias_variable([512])
conv6_3 = tf.nn.atrous_conv2d( conv6_2, W6_3, 2, padding = 'SAME') + b6_3
conv6_3 = tf.nn.relu(conv6_3)
conv6_3 = tf.contrib.layers.batch_norm(conv6_3)
W7_1 = weight_variable([3,3,512,256])
b7_1 = bias_variable([256])
conv7_1 = conv2d( conv6_3, W7_1, 1 ) + b7_1
conv7_1 = tf.nn.relu(conv7_1)
W7_2 = weight_variable([3,3,256,256])
b7_2 = bias_variable([256])
conv7_2 = conv2d( conv7_1, W7_2, 1 ) + b7_2
conv7_2 = tf.nn.relu(conv7_2)
W7_3 = weight_variable([3,3,256,256])
b7_3 = bias_variable([256])
conv7_3 = conv2d( conv7_2, W7_3, 1 ) + b7_3
conv7_3 = tf.nn.relu(conv7_3)
conv7_3 = tf.contrib.layers.batch_norm(conv7_3)
conv7_3 = tf.image.resize_images(conv7_3, [64,64])
W8_1 = weight_variable([4,4,256,128])
b8_1 = bias_variable([128])
conv8_1 = conv2d( conv7_3, W8_1, 1 ) + b8_1
conv8_1 = tf.nn.relu(conv8_1)
W8_2 = weight_variable([3,3,128,128])
b8_2 = bias_variable([128])
conv8_2 = conv2d( conv8_1, W8_2, 1 ) + b8_2
conv8_2 = tf.nn.relu(conv8_2)
W8_3 = weight_variable([3,3,128,128])
b8_3 = bias_variable([128])
conv8_3 = conv2d( conv8_2, W8_3, 1 ) + b8_3
conv8_3 = tf.nn.relu(conv8_3)
W_ab = weight_variable([1,1,128,313])
b_ab = bias_variable([313])
conv_ab = conv2d( conv8_3, W_ab, 1 ) + b_ab
output = tf.nn.relu(conv_ab)
return image_, output_, output
In [4]:
def loss_function(output, output_):
loss = tf.nn.softmax_cross_entropy_with_logits( output, output_ )
return tf.reduce_mean(loss)
In [5]:
def get_prediction( output ):
prediction = tf.nn.softmax( output )
return tf.image.resize_images( prediction, [256,256] )
In [6]:
image_, output_, output = setup_tensorflow_graph(1)
loss = loss_function(output, output_)
prediction = get_prediction(output)
In [7]:
prediction.get_shape()
Out[7]:
In [8]:
train_step = tf.train.AdamOptimizer(1e-4).minimize(loss)
sess = tf.Session()
sess.run(tf.initialize_all_variables())
sess.run( train_step, feed_dict = {image_ : np.ones((1,256,256,1)), output_ : np.ones((1,64,64,313))})