{ "cells": [ { "cell_type": "markdown", "metadata": { "_cell_guid": "2bf6e055-8947-53dd-023d-c2388b034968" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "4d1ef4e0-6ffc-1d30-18e8-67eb7a03af13" }, "outputs": [], "source": [ "import tensorflow as tf\n", "import numpy as np\n", "import matplotlib.pyplot as plt\n", "from tqdm import tqdm\n", "from IPython.display import display, Image, HTML\n", "import os\n", "import cv2\n", "\n", "TRAIN_DIR = '../input/train'\n", "TEST_DIR = '../input/test'\n", "IMG_SIZE = 64" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "09304d83-35a4-ae72-c0a9-e086e62f3431" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "90dd442c-918c-75c1-51fe-c47efd127943" }, "outputs": [], "source": [ "def create_input_data(im):\n", " img = cv2.imread(im, cv2.IMREAD_COLOR)\n", " if (img.shape[0] >= img.shape[1]): # height is greater than width\n", " resizeto = (IMG_SIZE, int (round (IMG_SIZE * (float (img.shape[1]) / img.shape[0]))));\n", " else:\n", " resizeto = (int (round (IMG_SIZE * (float (img.shape[0]) / img.shape[1]))), IMG_SIZE);\n", " \n", " img2 = cv2.resize(img, (resizeto[1], resizeto[0]), interpolation=cv2.INTER_CUBIC)\n", " img3 = cv2.copyMakeBorder(img2, 0, IMG_SIZE - img2.shape[0], 0, IMG_SIZE - img2.shape[1], cv2.BORDER_CONSTANT, 0)\n", " \n", " return img3[:,:,::-1]\n", "# img = cv2.imread(im, cv2.IMREAD_COLOR)\n", "# img = cv2.resize(img, (IMG_SIZE, IMG_SIZE))\n", "# return np.array(img / 255)" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "7dbc9749-607a-19a2-b55d-831912c97259" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "ab674bb9-bb6c-6ef4-08c1-c09e7c4f0573" }, "outputs": [], "source": [ "def one_hot_encode(img): \n", " if 'cat' in img:\n", " return np.array([0, 1])\n", " else:\n", " return np.array([1, 0])" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "73d4e49c-12d2-28be-2f92-89c9af19165d" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "735cf56f-826a-5bf1-45d1-294cecc26a3e" }, "outputs": [], "source": [ "training_img = []\n", "training_label = []\n", "\n", "testing_img = []\n", "testing_label = []\n", "\n", "for img in tqdm(os.listdir(TRAIN_DIR)):\n", " training_path = os.path.join(TRAIN_DIR, img)\n", " \n", " train_img = create_input_data(training_path)\n", " training_img.append(np.array(train_img))\n", " \n", " train_label = one_hot_encode(img)\n", " training_label.append(np.array(train_label))\n", " \n", "for img in tqdm(os.listdir(TEST_DIR)):\n", " testing_path = os.path.join(TEST_DIR, img)\n", " \n", " test_img = create_input_data(testing_path)\n", " testing_img.append(np.array(test_img))\n", " \n", " test_label = one_hot_encode(img)\n", " testing_label.append(np.array(test_label))\n", "\n", "training_img = np.array(training_img, dtype=np.float32)\n", "training_label = np.array(training_label, dtype=np.float32)\n", "\n", "testing_img = np.array(testing_img, dtype=np.float32)\n", "testing_label = np.array(testing_label, dtype=np.float32)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "b4df2a05-8f50-b6a0-a414-055388caaa94" }, "outputs": [], "source": [ "index = 15000\n", "plt.imshow(training_img[index])\n", "plt.show()\n", "print(training_label[index])" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "25b634c6-434e-25c8-fb98-243b0028fe75" }, "source": [ "### Input\n", "The neural network needs to read the image data, one-hot encoded labels, and dropout keep probability. Implement the following functions\n", "* Implement neural_net_image_input\n", " * Return a TF Placeholder \n", " * Set the shape using image_shape with batch size set to None.\n", " * Name the TensorFlow placeholder \"x\" using the TensorFlow name parameter in the TF Placeholder.\n", "* Implement neural_net_label_input\n", " * Return a TF Placeholder\n", " * Set the shape using n_classes with batch size set to None.\n", " * Name the TensorFlow placeholder \"y\" using the TensorFlow name parameter in the TF Placeholder.\n", "* Implement neural_net_keep_prob_input\n", " * Return a TF Placeholder for dropout keep probability.\n", " * Name the TensorFlow placeholder \"keep_prob\" using the TensorFlow name parameter in the TF Placeholder.\n", " * These names will be used at the end of the project to load your saved model.\n", "\n", "Note: _None_ for shapes in TensorFlow allow for a dynamic size." ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "bc429bf1-dc6a-6ba0-46d2-948e34013fd0" }, "outputs": [], "source": [ "def neural_net_image_input(image_size):\n", " return tf.placeholder(tf.float32, [None] + list(image_size), 'x')\n", "\n", "def neural_net_label_input():\n", " return tf.placeholder(tf.float32, [None, 2], 'y')\n", "\n", "def neural_net_keep_prob():\n", " return tf.placeholder(tf.float32, None, 'keep_prob')" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "94a5c19b-4beb-097f-a02a-96ee2aa8c088" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "01cf2987-73a5-adc6-242d-48fa9c47cfd4" }, "outputs": [], "source": [ "def conv2d_maxpool(x_tensor, conv_num_outputs, conv_ksize=(2,2), conv_strides=[1,1,1,1], pool_ksize=[1,2,2,1], pool_strides=[1,2,2,1]):\n", " dimension = x_tensor.get_shape().as_list()\n", " shape = list(conv_ksize + (dimension[-1],) + (conv_num_outputs,))\n", " weight = tf.Variable(tf.truncated_normal(shape, 0, 0.1))\n", " bias = tf.Variable(tf.zeros(conv_num_outputs))\n", " \n", " conv_layer = tf.nn.conv2d(x_tensor, weight, strides=conv_strides, padding='SAME')\n", " conv_layer = tf.nn.bias_add(conv_layer, bias)\n", " conv_layer = tf.nn.relu(conv_layer)\n", " \n", " conv_layer = tf.nn.max_pool(conv_layer, ksize=pool_ksize, strides=pool_strides, padding='SAME')\n", " \n", " return conv_layer" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "339f750c-1f1e-3476-0d02-a7c548be0ef9" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "93a04505-c9a3-4e6c-a69d-b26684ab2188" }, "outputs": [], "source": [ "def flatten(x_tensor):\n", " dimension = x_tensor.get_shape().as_list()\n", " return tf.reshape(x_tensor, [-1, np.prod(dimension[1:])])" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "afbd2dfd-3bba-d3f0-0c34-1605616bd72d", "collapsed": true }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "9e5e5835-d795-bcc0-f55b-718018e6e92a" }, "outputs": [], "source": [ "def fully_conn(x_tensor, num_outputs):\n", " dimension = x_tensor.get_shape().as_list()\n", " shape = list((dimension[-1],) + (num_outputs,))\n", " weights = tf.Variable(tf.truncated_normal(shape, 0, 0.1))\n", " bias = tf.Variable(tf.zeros(num_outputs))\n", " \n", " fully_conn = tf.nn.relu(tf.add(tf.matmul(x_tensor, weights), bias))\n", " \n", " return fully_conn" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "c88633fb-1834-8712-4904-3b3c4d8d3e6b" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "3c9ea56a-5bde-06b3-889d-51975bb66a1b" }, "outputs": [], "source": [ "def output(x_tensor, num_outputs):\n", " dimension = x_tensor.get_shape().as_list()\n", " shape = list((dimension[-1],) + (num_outputs,))\n", " weights = tf.Variable(tf.truncated_normal(shape, 0, 0.01))\n", " bias = tf.Variable(tf.zeros(num_outputs))\n", " \n", " output = tf.add(tf.matmul(x_tensor, weights), bias)\n", " \n", " return output" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "d432bf0f-6928-b568-e736-fabb549d25d9" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "b67b5ac7-9730-c1c1-d1a8-687d2e724eb0" }, "outputs": [], "source": [ "def conv_net(x, keep_prob):\n", " \n", " model = conv2d_maxpool(x, conv_num_outputs=32) \n", " model = tf.nn.dropout(model, keep_prob)\n", " \n", " model = conv2d_maxpool(x, conv_num_outputs=64) \n", " model = tf.nn.dropout(model, keep_prob)\n", " \n", " model = flatten(model)\n", " model = tf.nn.dropout(model, keep_prob)\n", " \n", " model = fully_conn(model, 128)\n", " \n", " model = output(model, 2)\n", " \n", " return model\n", "\n", "##############################\n", "## Build the Neural Network ##\n", "##############################\n", "\n", "# Remove previous weights, bias, inputs, etc..\n", "tf.reset_default_graph()\n", "\n", "# Inputs\n", "x = neural_net_image_input((IMG_SIZE, IMG_SIZE, 3))\n", "y = neural_net_label_input()\n", "keep_prob = neural_net_keep_prob()\n", "\n", "# Model\n", "logits = conv_net(x, keep_prob)\n", "\n", "# Name logits Tensor, so that is can be loaded from disk after training\n", "logits = tf.identity(logits, name='logits')\n", "\n", "# Loss and Optimizer\n", "cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=y))\n", "optimizer = tf.train.AdamOptimizer(1e-3).minimize(cost)\n", "\n", "# Accuracy\n", "correct_pred = tf.equal(tf.argmax(logits, 1), tf.argmax(y, 1))\n", "accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32), name='accuracy')" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "91da20e0-ebc1-a6e1-c3fc-f48bc5f4f635" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "b7681600-bf2f-2c61-43db-420472996694" }, "outputs": [], "source": [ "def train_neural_network(session, optimizer, keep_probability, feature_batch, label_batch):\n", " session.run(optimizer, feed_dict={x:feature_batch, y:label_batch, keep_prob:keep_probability})" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "44a0a8d2-c6b6-9882-85e1-43db8702ae32" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "ca0a4c4e-be32-c55a-2ff8-85135b8dfef8" }, "outputs": [], "source": [ "def print_stats(session, feature_batch, label_batch, cost, accuracy):\n", " loss = session.run(cost, feed_dict={x:feature_batch, y:label_batch, keep_prob:0.7})\n", " valid_acc = sess.run(accuracy, feed_dict={\n", " x: training_img[:batch_size],\n", " y: training_label[:batch_size],\n", " keep_prob: 0.7})\n", " print('Loss: {:>10.4f} Validation Accuracy: {:.6f}'.format(loss, valid_acc))" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "3e02e280-00cd-f25f-e9c9-d0bbb405edb8" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "5f929689-943b-b6c7-1974-0f90687a1283" }, "outputs": [], "source": [ "# TODO: Tune Parameters\n", "epochs = 10\n", "batch_size = 64\n", "keep_probability = 0.7" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "a88209f9-7aab-b720-8eb6-ed57c97093b5" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "f3e4b115-d179-d1d8-5c9e-56895b1cf987" }, "outputs": [], "source": "" }, { "cell_type": "markdown", "metadata": { "_cell_guid": "66e5851b-beb1-ff11-cf7a-cf8e4ea9690f" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "03834202-47ac-eb5f-2c5d-1f44a9d299f8" }, "outputs": [], "source": [ "save_model_path = './image_classification'\n", "\n", "print('Training...')\n", "with tf.Session() as sess:\n", " # Initializing the variables\n", " sess.run(tf.global_variables_initializer())\n", " \n", " # Training cycle\n", " for epoch in range(epochs):\n", " # Loop over all batches\n", " n_batches = 5\n", " for batch_i in range(1, n_batches + 1):\n", " batch_features = training_img[batch_i:batch_size]\n", " batch_labels = training_label[batch_i:batch_size]\n", " \n", " train_neural_network(sess, optimizer, keep_probability, batch_features, batch_labels)\n", " \n", " print('Epoch {:>2}, Batch {}: '.format(epoch + 1, batch_i), end='')\n", " print_stats(sess, batch_features, batch_labels, cost, accuracy)\n", "\n", " # Save Model\n", " saver = tf.train.Saver()\n", " save_path = saver.save(sess, save_model_path)" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "1d5a081d-f2f3-3a1b-2cc4-547297b29e91" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "b01a6962-87d2-d64f-a215-bcd5d59d55ee" }, "outputs": [], "source": [ "# Set batch size if not already set\n", "try:\n", " if batch_size:\n", " pass\n", "except NameError:\n", " batch_size = 64\n", "\n", "save_model_path = './image_classification'\n", "n_samples = 4\n", "top_n_predictions = 3\n", "\n", "def test_model():\n", " \"\"\"\n", " Test the saved model against the test dataset\n", " \"\"\"\n", "\n", "# test_features, test_labels = pickle.load(open('preprocess_training.p', mode='rb'))\n", " loaded_graph = tf.Graph()\n", "\n", " with tf.Session(graph=loaded_graph) as sess:\n", " # Load model\n", " loader = tf.train.import_meta_graph(save_model_path + '.meta')\n", " loader.restore(sess, save_model_path)\n", "\n", " # Get Tensors from loaded model\n", " loaded_x = loaded_graph.get_tensor_by_name('x:0')\n", " loaded_y = loaded_graph.get_tensor_by_name('y:0')\n", " loaded_keep_prob = loaded_graph.get_tensor_by_name('keep_prob:0')\n", " loaded_logits = loaded_graph.get_tensor_by_name('logits:0')\n", " loaded_acc = loaded_graph.get_tensor_by_name('accuracy:0')\n", " \n", " test_epoch = 25\n", " # Get accuracy in batches for memory limitations\n", " test_batch_acc_total = 0\n", " test_batch_count = 0\n", " \n", " n_batches = 500\n", " b_size = 0\n", " for i in range(25):\n", " test_batch_acc_total += sess.run(\n", " loaded_acc,\n", " feed_dict={loaded_x: testing_img[b_size:n_batches], loaded_y: testing_label[b_size:n_batches], loaded_keep_prob: 1.0})\n", " test_batch_count += 1 \n", "\n", " print('Batch {:>2}: Testing Accuracy: {}\\n'.format(i + 1, test_batch_acc_total/test_batch_count), end='')\n", " \n", " b_size = n_batches + 1\n", " n_batches += 500\n", "\n", "\n", "test_model()" ] } ], "metadata": { "_change_revision": 0, "_is_fork": false, "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.6.0" } }, "nbformat": 4, "nbformat_minor": 0 }