Simple 2-layer MLP to clarify a TensorFlow API question. Uses the StatOil/C-CORE dataset from Kaggle.

statoil_mlp.py

import numpy as np
import pandas as pd
import tensorflow as tf

def normalize(arr, axis):
    means = arr.mean(axis=axis, keepdims=True)
    devs = arr.std(axis=axis, keepdims=True)
    arr -= means
    arr /= devs    

df = pd.read_json('data/processed/train.json')
images = np.array([df['band_1'], df['band_2']], dtype = np.float32)
images = np.moveaxis(images, 0, -1).reshape(-1, 75, 75, 2)
labels = df['is_iceberg'].as_matrix()
normalize(images, axis=(0, 1, 2))

seed = 42
tf.set_random_seed(seed)
np.random.seed(seed)
tf.reset_default_graph()

X = tf.placeholder(tf.float32, shape=(None, 75, 75, 2))
y = tf.placeholder(tf.int32, shape=(None))    
fc0 = tf.layers.flatten(X)
fc1 = tf.layers.dense(fc0, 512, activation=tf.nn.relu)
fc2 = tf.layers.dense(fc1, 256, activation=tf.nn.relu)
# Setup 1:
out = tf.layers.dense(fc2, 2)
loss = tf.losses.sparse_softmax_cross_entropy(labels=y, logits=out)
correct_flags = tf.nn.in_top_k(out, y, 1)
# Setup 2:
# out = tf.layers.dense(fc2, 1, activation=tf.sigmoid)
# loss = tf.losses.log_loss(labels=y, predictions=out)
# correct_flags = tf.equal(y, tf.cast(tf.round(out), tf.int32))

train_step = tf.train.AdamOptimizer(0.001).minimize(loss)
accuracy = tf.reduce_mean(tf.cast(correct_flags, tf.float32))    

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    for e in range(100):
        _, loss_val, acc_val = sess.run([train_step, loss, accuracy], feed_dict={X: images, y: labels})
        print('Epoch: {} Loss: {} Accuracy: {}'.format(e + 1, loss_val, acc_val))