FCN-trainer-improved-val.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model Loaded\n",
      "conv1_1 Value Loaded\n",
      "Layer name: conv1_1\n",
      "Layer shape: (3, 3, 3, 64)\n",
      "conv1_1 Frozen\n",
      "conv1_1 (64,) Value Loaded\n",
      "\n",
      "\n",
      "conv1_2 Value Loaded\n",
      "Layer name: conv1_2\n",
      "Layer shape: (3, 3, 64, 64)\n",
      "conv1_2 Frozen\n",
      "conv1_2 (64,) Value Loaded\n",
      "\n",
      "\n",
      "conv2_1 Value Loaded\n",
      "Layer name: conv2_1\n",
      "Layer shape: (3, 3, 64, 128)\n",
      "L2 Loss Added\n",
      "conv2_1 (128,) Value Loaded\n",
      "\n",
      "\n",
      "conv2_2 Value Loaded\n",
      "Layer name: conv2_2\n",
      "Layer shape: (3, 3, 128, 128)\n",
      "L2 Loss Added\n",
      "conv2_2 (128,) Value Loaded\n",
      "\n",
      "\n",
      "conv3_1 Value Loaded\n",
      "Layer name: conv3_1\n",
      "Layer shape: (3, 3, 128, 256)\n",
      "L2 Loss Added\n",
      "conv3_1 (256,) Value Loaded\n",
      "\n",
      "\n",
      "conv3_2 Value Loaded\n",
      "Layer name: conv3_2\n",
      "Layer shape: (3, 3, 256, 256)\n",
      "L2 Loss Added\n",
      "conv3_2 (256,) Value Loaded\n",
      "\n",
      "\n",
      "conv3_3 Value Loaded\n",
      "Layer name: conv3_3\n",
      "Layer shape: (3, 3, 256, 256)\n",
      "L2 Loss Added\n",
      "conv3_3 (256,) Value Loaded\n",
      "\n",
      "\n",
      "conv4_1 Value Loaded\n",
      "Layer name: conv4_1\n",
      "Layer shape: (3, 3, 256, 512)\n",
      "L2 Loss Added\n",
      "conv4_1 (512,) Value Loaded\n",
      "\n",
      "\n",
      "conv4_2 Value Loaded\n",
      "Layer name: conv4_2\n",
      "Layer shape: (3, 3, 512, 512)\n",
      "L2 Loss Added\n",
      "conv4_2 (512,) Value Loaded\n",
      "\n",
      "\n",
      "conv4_3 Value Loaded\n",
      "Layer name: conv4_3\n",
      "Layer shape: (3, 3, 512, 512)\n",
      "L2 Loss Added\n",
      "conv4_3 (512,) Value Loaded\n",
      "\n",
      "\n",
      "conv5_1 Value Loaded\n",
      "Layer name: conv5_1\n",
      "Layer shape: (3, 3, 512, 512)\n",
      "L2 Loss Added\n",
      "conv5_1 (512,) Value Loaded\n",
      "\n",
      "\n",
      "conv5_2 Value Loaded\n",
      "Layer name: conv5_2\n",
      "Layer shape: (3, 3, 512, 512)\n",
      "L2 Loss Added\n",
      "conv5_2 (512,) Value Loaded\n",
      "\n",
      "\n",
      "conv5_3 Value Loaded\n",
      "Layer name: conv5_3\n",
      "Layer shape: (3, 3, 512, 512)\n",
      "L2 Loss Added\n",
      "conv5_3 (512,) Value Loaded\n",
      "\n",
      "\n",
      "Layer name: fc6\n",
      "Layer shape: [7, 7, 512, 4096]\n",
      "fc6 Value Loaded\n",
      "fc6 (4096,) Value Loaded\n",
      "\n",
      "\n",
      "Layer name: fc7\n",
      "Layer shape: [1, 1, 4096, 4096]\n",
      "fc7 Value Loaded\n",
      "fc7 (4096,) Value Loaded\n",
      "\n",
      "\n",
      "fc8 Value Randomly initialized\n",
      "L2 Loss Added\n",
      "Layer name: score_fr\n",
      "Layer shape: [1, 1, 4096, 19]\n",
      "fc8_biases Value Initialized 0\n",
      "\n",
      "\n",
      "Finished building Network.\n",
      "using Pretrained model\n",
      "model_checkpoint_path: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-81\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-64\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-65\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-66\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-67\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-68\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-69\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-70\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-71\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-72\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-73\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-74\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-75\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-76\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-77\"\n",
      "all_model_checkpoint_paths: \"/mnt/data1/fcn/ckpt_dir/model.ckpt-81\"\n",
      "\n",
      "finished\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib as mpl\n",
    "import matplotlib.cm as cm\n",
    "import copy\n",
    "import tensorflow as tf\n",
    "\n",
    "import os\n",
    "import time\n",
    "import numpy as np\n",
    "from IPython.display import clear_output\n",
    "import sys\n",
    "import csv\n",
    "import math\n",
    "from matplotlib import colors\n",
    "import fcn32_new as fcn32_vgg\n",
    "import csv\n",
    "ckpt_dir = \"/mnt/data1/fcn/ckpt_dir\"\n",
    "PRETRAINED_MODEL_PATH= \"/mnt/data1/fcn/ckpt_dir\"\n",
    "LOGS_PATH = '/home/sik4hi/tensorflow_logs'\n",
    "WEIGHT_PATH = '.npy'\n",
    "TRAINSET_PATH = '/mnt/data1/city/csv_files/cityscapes_train_wmask.csv'\n",
    "VALSET_PATH = '/mnt/data1/city/csv_files/cityscapes_val_wmask.csv'\n",
    "\n",
    "NUM_OF_CLASSESS = 19\n",
    "BATCH_SIZE = 1\n",
    "IMAGE_HEIGHT = 1024\n",
    "IMAGE_WIDTH = 2048\n",
    "NUM_CHANNELS = 3\n",
    "N_EPOCHS = 1\n",
    "\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"1\"\n",
    "f=open(VALSET_PATH,\"r\")\n",
    "filepaths=[]\n",
    "labels=[]\n",
    "masks=[]\n",
    "for line in f:\n",
    "    filepath, label, mask= line.split(\",\")\n",
    "    filepaths.append(filepath)\n",
    "    labels.append(label)\n",
    "    masks.append(mask.split(\"\\r\")[0])\n",
    "    \n",
    "one_ex = tf.train.slice_input_producer([filepaths,labels, masks]\n",
    "                                           ,capacity = 500)\n",
    "#csv_path = tf.train.string_input_producer([TRAINSET_PATH])\n",
    "#textReader = tf.TextLineReader()\n",
    "#_, content = textReader.read(csv_path)\n",
    "#one_ex = tf.decode_csv(content, record_defaults=[[\"\"], [\"\"], [\"\"]])\n",
    "\n",
    "im_content = tf.read_file(one_ex[0])\n",
    "train_image = tf.image.decode_png(im_content, channels=3)\n",
    "train_image.set_shape([IMAGE_HEIGHT, IMAGE_WIDTH, NUM_CHANNELS])\n",
    "train_image = tf.cast(train_image, tf.float32)\n",
    "\n",
    "la_content = tf.read_file(one_ex[1])\n",
    "label_image = tf.image.decode_png(la_content, channels=1)\n",
    "label_image.set_shape([IMAGE_HEIGHT, IMAGE_WIDTH, 1])\n",
    "label_image=tf.squeeze(label_image, squeeze_dims=[2])\n",
    "\n",
    "#record_bytes=2097152\n",
    "#reader=tf.FixedLengthRecordReader(record_bytes)\n",
    "value =tf.read_file(one_ex[2])\n",
    "mask=tf.decode_raw(value,tf.uint8)\n",
    "mask.set_shape([2097152])\n",
    "#mask=tf.cast(mask, tf.float32)\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "train_image_batch, train_label_batch, train_mask_batch= tf.train.batch([train_image, label_image, mask], batch_size=BATCH_SIZE,\n",
    "                                                               capacity=100 + 3 * BATCH_SIZE,\n",
    "                                                               num_threads=7)\n",
    "with tf.device('/gpu:0'):\n",
    "    sess = tf.Session()\n",
    "    #images_tf = tf.placeholder(tf.float32,[None, 1024, 2048, 3])\n",
    "    #labels_tf = tf.placeholder(tf.int32,[None, 1024, 2048])\n",
    "    #weights = tf.placeholder(tf.float32,[None, 1024, 2048])\n",
    "\n",
    "    vgg_fcn = fcn32_vgg.FCN32VGG('./vgg16.npy')\n",
    "    with tf.name_scope(\"content_vgg\"):\n",
    "        vgg_fcn.build(train_image_batch, train=False, num_classes=19, random_init_fc8=True, debug=False)\n",
    "\n",
    "\n",
    "    #head=[]\n",
    "    #sum_of_weights=tf.reduce_sum(weights)\n",
    "    #labels_oh= tf.one_hot(labels_tf,19)\n",
    "    labels_tf_flat=tf.reshape(tf.mul(tf.reshape(train_label_batch,[-1]),train_mask_batch),[-1])\n",
    "    mask2=tf.reshape(train_mask_batch,[1024,2048])\n",
    "    #labels_tf_flat2= tf.gather(labels_tf_flat,tf.where(tf.not_equal(labels_tf_flat,19))[1])\n",
    "    logits = tf.reshape(vgg_fcn.upscore, (-1, NUM_OF_CLASSESS))\n",
    "    #logits = tf.gather(logits,tf.where(tf.not_equal(labels_tf_flat,19)))\n",
    "    labels_tf_flat = tf.cast(labels_tf_flat, tf.int32)\n",
    "\n",
    "    #epsilon = tf.constant(value=1e-4)\n",
    "    #logits = logits + epsilon\n",
    "    #labels_flat_sparse = tf.reshape(labels_oh, (-1, NUM_OF_CLASSESS))\n",
    "    #labels_flat_sparse=tf.cast(labels_flat_sparse, tf.float64)\n",
    "    #ww_flat=tf.reshape(weights,[-1])\n",
    "    #ww_flat = tf.cast(ww_flat, tf.float64)\n",
    "    #softmax = tf.nn.softmax(logits)\n",
    "    #softmax = tf.cast(softmax, tf.float64)\n",
    "    #test1=-tf.reduce_sum(labels_flat_sparse * tf.log(softmax), reduction_indices=[1])\n",
    "    #test2=tf.reduce_sum(test1)\n",
    "    #cross_entropy = tf.mul(-tf.reduce_sum(labels_flat_sparse * tf.log(softmax), reduction_indices=[1]),ww_flat)\n",
    "    #cross_entropy = -tf.reduce_sum(labels_flat_sparse * tf.log(softmax), reduction_indices=[1])\n",
    "    #mask_flat=tf.reshape(train_mask_batch,[-1])\n",
    "    train_mask_batch=tf.cast(train_mask_batch, tf.float32)\n",
    "    cross_entropy =tf.mul(tf.nn.sparse_softmax_cross_entropy_with_logits(logits,labels_tf_flat),train_mask_batch)\n",
    "    #cross_entropy_sum= tf.reduce_sum(cross_entropy,\n",
    "    #                                name='xentropy_mean')\n",
    "    #cross_entropy_mean = tf.reduce_mean(cross_entropy,name='xentropy_mean')\n",
    "    #cross_entropy_mean = cross_entropy_sum/tf.cast(sum_of_weights, tf.float32)\n",
    "    cross_entropy_mean = tf.reduce_mean(cross_entropy,\n",
    "                                        name='xentropy_mean')\n",
    "    l2_loss=tf.reduce_sum(tf.get_collection(\"losses\"))\n",
    "    total_loss=cross_entropy_mean +l2_loss\n",
    "    #loss_tf = tf.reduce_mean((tf.nn.sparse_softmax_cross_entropy_with_logits(vgg_fcn.upscore,\n",
    "     #                                                                     tf.squeeze(labels_tf, squeeze_dims=[3]),\n",
    "      #                                                                    name=\"entropy\")))\n",
    "    \n",
    "    \n",
    "    \n",
    "    \n",
    "    #train_op = tf.train.MomentumOptimizer(0.0001, 0.99).minimize(cross_entropy_mean)\n",
    "    #train_op = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(total_loss)\n",
    "    predictions=tf.argmax(logits, 1)\n",
    "    #labels_tf_flat=tf.reshape(labels_tf_flat,[-1])\n",
    "    #labels_flat=tf.cast(labels_tf_flat, tf.int64)\n",
    "    pred_flat=tf.reshape(predictions,[-1])\n",
    "with tf.device('/cpu:0'):\n",
    "    miou=tf.contrib.metrics.streaming_mean_iou(pred_flat,labels_tf_flat,19,weights=train_mask_batch)\n",
    "    if not os.path.exists(ckpt_dir):\n",
    "        os.makedirs(ckpt_dir)\n",
    "    saver = tf.train.Saver(max_to_keep=10)\n",
    "with tf.device('/gpu:0'):\n",
    "    print('Finished building Network.')\n",
    "    \n",
    "    init_op = tf.group(tf.initialize_all_variables(),\n",
    "                       tf.initialize_local_variables())\n",
    "    sess.run(init_op)\n",
    "        \n",
    "    if PRETRAINED_MODEL_PATH:\n",
    "        print \"using Pretrained model\"\n",
    "        ckpt = tf.train.get_checkpoint_state(PRETRAINED_MODEL_PATH)\n",
    "        print(ckpt)\n",
    "        saver.restore(sess,ckpt.model_checkpoint_path)\n",
    "        \n",
    "    # print(csv_path)\n",
    "    # For populating queues with batches, very important!\n",
    "    coord = tf.train.Coordinator()\n",
    "    threads = tf.train.start_queue_runners(sess=sess,coord=coord)\n",
    "    \n",
    "    print(\"finished\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "epoch = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===========**Training ACCURACY**================\n",
      "Epoch 2\n",
      "Training mIOU:  0.457574\n",
      "Training Loss: 0.264172\n",
      "Training Loss: 0.326454\n",
      "Time Elapsed for Epoch:1 is 2.95912611485 minutes\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAqoAAAFwCAYAAABwyFm+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSQZVSZIkVdL/D7KCgHl0sQtZAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe0c00c1ed0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAqoAAAFwCAYAAABwyFm+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nVZIkSUUyqEqSJKlI/x9nWzvUWcddJgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe08c701b10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAqEAAAFwCAYAAACIKKJJAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nhuSstOKKAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe08c69d4d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "requesting stop\n"
     ]
    }
   ],
   "source": [
    "total_train_loss_list,train_loss_list,plot_train_loss,plot_total_train_loss,plot_miou,loss= [],[],[],[],[],0\n",
    "#for epoch in xrange(N_EPOCHS):\n",
    "cmap = colors.ListedColormap(['blue','lightblue','orange','purple','r'\n",
    "                              ,'darkkhaki','yellow','saddlebrown','orchid','grey'\n",
    "                              ,'hotpink','blueviolet','gold','deeppink','greenyellow'\n",
    "                              ,'maroon','yellowgreen','green','orangered','black'])\n",
    "norm=colors.NoNorm(0,19)\n",
    "try:\n",
    "    while not coord.should_stop() and epoch< N_EPOCHS:\n",
    "        epoch+=1\n",
    "        if(math.isnan(loss)):\n",
    "                break\n",
    "        epoch_start_time = time.time()\n",
    "        for iteration in xrange(500/BATCH_SIZE):   \n",
    "            #train_imbatch, train_labatch = sess.run([train_image_batch, train_label_batch])\n",
    "            #image1=copy.deepcopy(train_labatch[0])\n",
    "            #ww=[[[0 for k in range(2048)] for j in range(1024)]for i in range(BATCH_SIZE)]\n",
    "            #sum2=0\n",
    "            #for i in range(BATCH_SIZE):\n",
    "            #    for j in range(1024):\n",
    "            #        for k in range(2048):\n",
    "            #            if(train_labatch[i][j][k]==19):\n",
    "            #                ww[i][j][k]=0\n",
    "            #                train_labatch[i][j][k]=18\n",
    "            #            else:\n",
    "            #                ww[i][j][k]=1\n",
    "                            #sum2+=1\n",
    "           # sum2=sess.run([sum_of_weights],\n",
    "           #         feed_dict={weights: ww})\n",
    "            if (iteration%100==0):\n",
    "                loss,tloss,iou,output, output3 = sess.run([cross_entropy_mean,total_loss,\n",
    "                                                            miou,vgg_fcn.pred_up,train_label_batch])\n",
    "                plt.figure(1)\n",
    "                plt.imshow(output3[0],cmap=cmap,norm=norm)\n",
    "                plt.figure(2)\n",
    "                plt.imshow(output[0],cmap=cmap,norm=norm)\n",
    "                #plt.figure(3)\n",
    "                #plt.imshow(output2[0],cmap=cmap,norm=norm)\n",
    "                plt.show()\n",
    "                \n",
    "            else:\n",
    "                loss,tloss,iou,output,output2 = sess.run([cross_entropy_mean,total_loss,\n",
    "                                                            miou,vgg_fcn.pred_up,vgg_fcn.pred])\n",
    "\n",
    "            if(math.isnan(loss)):\n",
    "                break\n",
    "            train_loss_list.append(loss)\n",
    "            total_train_loss_list.append(tloss)\n",
    "            sys.stdout.write('\\r\\r\\r\\r\\r\\r\\r\\r\\r\\r' + \"Iteration:\" + str(iteration)+ \"  Loss: \"+ str(loss)+ \" Total Loss: \"+ str(tloss))\n",
    "            #print(loss)\n",
    "            #print(val1)\n",
    "            #print(val2)\n",
    "            #print sum2\n",
    "            sys.stdout.flush()\n",
    "\n",
    "            #if (iteration%500==0):\n",
    "                \n",
    "\n",
    "\n",
    "        tl = np.mean(train_loss_list)\n",
    "        ttl=np.mean(total_train_loss_list)\n",
    "        plot_train_loss.append(tl)\n",
    "        plot_total_train_loss.append(ttl)\n",
    "\n",
    "        train_loss_list = []\n",
    "        total_train_loss_list = []\n",
    "        plot_miou.append(iou[0])\n",
    "\n",
    "        clear_output()\n",
    "        print \"===========**Training ACCURACY**================\"\n",
    "        print \"Epoch\", epoch + 1\n",
    "        print 'Training mIOU: ', iou[0]\n",
    "        print \"Training Loss:\", tl \n",
    "        print \"Training Loss:\", ttl \n",
    "        print 'Time Elapsed for Epoch:' + str(epoch) + ' is ' + str(\n",
    "                (time.time() - epoch_start_time) / 60.) + ' minutes'\n",
    "\n",
    "        plt.figure(1) \n",
    "        dd = plt.plot(plot_total_train_loss,'r',label=\"Training\")\n",
    "        plt.title(\"Total LOSS\")\n",
    "        plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)\n",
    "        plt.figure(2) \n",
    "        aa = plt.plot(plot_train_loss,'r',label=\"Training\")\n",
    "        plt.title(\"LOSS\")\n",
    "        plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)\n",
    "        plt.figure(3)\n",
    "        cc = plt.plot(plot_miou,'r',label=\"Training\")\n",
    "        plt.title(\"mIOU\")\n",
    "        plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)\n",
    "        plt.show()\n",
    "        #saver.save(sess, ckpt_dir + \"/model.ckpt\", global_step=epoch)\n",
    "        ofile  = open('data.csv', \"a\")\n",
    "        writer = csv.writer(ofile)\n",
    "        timep=((time.time() - epoch_start_time) / 60.)\n",
    "        writer.writerow([tl,ttl,iou[0],timep,epoch,1])\n",
    "        ofile.close()\n",
    "except tf.errors.OutOfRangeError:\n",
    "    print (\"out of range\")\n",
    "finally:\n",
    "    coord.request_stop()\n",
    "    coord.join(threads)\n",
    "    print(\"requesting stop\")\n",
    "        #print(iou[1])\n",
    "            #train_loss= sess.run(vgg_fcn.upscore,\n",
    "             #   feed_dict={ images_tf: train_imbatch})\n",
    "            #res=sess.run(image)\n",
    "            #plt.imshow(train_labatch[0])\n",
    "            #plt.show()\n",
    "    #    aa= tt * np.log10(train_loss)\n",
    "    #    print(len(aa))\n",
    "    #    print(len(aa[0]))\n",
    "    #print(sum)\n",
    "    #print(sum2)\n",
    "    #print(len(ww))\n",
    "    #print(len(ww[0]))\n",
    "    #print(len(ww[0][0]))\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "cmap = colors.ListedColormap(['blue','purple','orange','lightblue','r'\n",
    "                              ,'darkkhaki','yellow','green','orchid','grey'\n",
    "                              ,'hotpink','blueviolet','gold','deeppink','greenyellow'\n",
    "                              ,'maroon','yellowgreen','saddlebrown','orangered','black'])\n",
    "norm=colors.NoNorm(0,19)\n",
    "output,output2,log= sess.run([train_label_batch,mask2,logits])\n",
    "print(len(log))\n",
    "plt.figure(1)\n",
    "plt.imshow(output[0],cmap=cmap,norm=norm)\n",
    "plt.figure(3)\n",
    "plt.imshow(output2,cmap=cmap,norm=norm)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "print(train_labatch[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}