BiSeNet_Road_Segmentation.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "BiSeNet - Road Segmentation.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/parasdahiya/BiSeNet-Road-Segmentation/blob/master/BiSeNet_Road_Segmentation.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Axm6b-VAJs6h",
        "colab_type": "code",
        "outputId": "1ad2333c-2a97-40f3-c955-8f2222767f5a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 125
        }
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Go to this URL in a browser: https://accounts.google.com/o/oauth2/auth?client_id=947318989803-6bn6qk8qdgf4n4g3pfee6491hc0brc4i.apps.googleusercontent.com&redirect_uri=urn%3Aietf%3Awg%3Aoauth%3A2.0%3Aoob&scope=email%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fdocs.test%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fdrive%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fdrive.photos.readonly%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fpeopleapi.readonly&response_type=code\n",
            "\n",
            "Enter your authorization code:\n",
            "··········\n",
            "Mounted at /content/gdrive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "idgoaSMZKJZs",
        "colab_type": "code",
        "outputId": "20f4de26-dddd-4db6-f744-f4e60e71a3fa",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "%cd /content/gdrive/'My Drive'/SemanticSegmentation"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/gdrive/My Drive/SemanticSegmentation\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YHZOR6pRAqxM",
        "colab_type": "code",
        "outputId": "a8bd9b90-4f3d-4a6f-caef-18604b1fdcd7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "!python train.py --ckpt"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING: Logging before flag parsing goes to stderr.\n",
            "W0906 11:39:45.255640 140289655293824 deprecation_wrapper.py:119] From /content/gdrive/My Drive/SemanticSegmentation/bisenet.py:74: The name tf.reset_default_graph is deprecated. Please use tf.compat.v1.reset_default_graph instead.\n",
            "\n",
            "W0906 11:39:45.255918 140289655293824 deprecation_wrapper.py:119] From /content/gdrive/My Drive/SemanticSegmentation/bisenet.py:78: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
            "\n",
            "W0906 11:39:45.258407 140289655293824 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:1251: calling VarianceScaling.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
            "Tensor(\"input_im:0\", shape=(?, ?, ?, 3), dtype=float32)\n",
            "Tensor(\"conv2d_8/BiasAdd:0\", shape=(?, ?, ?, 2), dtype=float32)\n",
            "W0906 11:39:51.998024 140289655293824 deprecation_wrapper.py:119] From train.py:30: The name tf.losses.softmax_cross_entropy is deprecated. Please use tf.compat.v1.losses.softmax_cross_entropy instead.\n",
            "\n",
            "W0906 11:39:52.029755 140289655293824 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/losses/losses_impl.py:121: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
            "W0906 11:39:52.040697 140289655293824 deprecation_wrapper.py:119] From train.py:32: The name tf.train.RMSPropOptimizer is deprecated. Please use tf.compat.v1.train.RMSPropOptimizer instead.\n",
            "\n",
            "W0906 11:39:54.679105 140289655293824 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/rmsprop.py:119: calling Ones.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
            "W0906 11:39:59.330440 140289655293824 deprecation_wrapper.py:119] From train.py:34: The name tf.train.Saver is deprecated. Please use tf.compat.v1.train.Saver instead.\n",
            "\n",
            "W0906 11:40:00.586906 140289655293824 deprecation_wrapper.py:119] From train.py:35: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.\n",
            "\n",
            "2019-09-06 11:40:00.588308: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcuda.so.1\n",
            "2019-09-06 11:40:00.613183: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.613974: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties: \n",
            "name: Tesla K80 major: 3 minor: 7 memoryClockRate(GHz): 0.8235\n",
            "pciBusID: 0000:00:04.0\n",
            "2019-09-06 11:40:00.614275: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudart.so.10.0\n",
            "2019-09-06 11:40:00.615722: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcublas.so.10.0\n",
            "2019-09-06 11:40:00.617515: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcufft.so.10.0\n",
            "2019-09-06 11:40:00.617929: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcurand.so.10.0\n",
            "2019-09-06 11:40:00.619776: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcusolver.so.10.0\n",
            "2019-09-06 11:40:00.621498: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcusparse.so.10.0\n",
            "2019-09-06 11:40:00.625757: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudnn.so.7\n",
            "2019-09-06 11:40:00.625943: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.626742: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.627489: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0\n",
            "2019-09-06 11:40:00.633252: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2300000000 Hz\n",
            "2019-09-06 11:40:00.633522: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x206e840 executing computations on platform Host. Devices:\n",
            "2019-09-06 11:40:00.633565: I tensorflow/compiler/xla/service/service.cc:175]   StreamExecutor device (0): <undefined>, <undefined>\n",
            "2019-09-06 11:40:00.689850: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.690743: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x206f480 executing computations on platform CUDA. Devices:\n",
            "2019-09-06 11:40:00.690776: I tensorflow/compiler/xla/service/service.cc:175]   StreamExecutor device (0): Tesla K80, Compute Capability 3.7\n",
            "2019-09-06 11:40:00.691006: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.691736: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties: \n",
            "name: Tesla K80 major: 3 minor: 7 memoryClockRate(GHz): 0.8235\n",
            "pciBusID: 0000:00:04.0\n",
            "2019-09-06 11:40:00.691796: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudart.so.10.0\n",
            "2019-09-06 11:40:00.691823: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcublas.so.10.0\n",
            "2019-09-06 11:40:00.691845: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcufft.so.10.0\n",
            "2019-09-06 11:40:00.691868: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcurand.so.10.0\n",
            "2019-09-06 11:40:00.691890: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcusolver.so.10.0\n",
            "2019-09-06 11:40:00.691912: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcusparse.so.10.0\n",
            "2019-09-06 11:40:00.691935: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudnn.so.7\n",
            "2019-09-06 11:40:00.692022: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.692771: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.693475: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0\n",
            "2019-09-06 11:40:00.693539: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudart.so.10.0\n",
            "2019-09-06 11:40:00.694989: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:\n",
            "2019-09-06 11:40:00.695019: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187]      0 \n",
            "2019-09-06 11:40:00.695032: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1200] 0:   N \n",
            "2019-09-06 11:40:00.695206: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.695947: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1005] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
            "2019-09-06 11:40:00.696637: W tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:40] Overriding allow_growth setting because the TF_FORCE_GPU_ALLOW_GROWTH environment variable is set. Original config value was 0.\n",
            "2019-09-06 11:40:00.696681: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1326] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10805 MB memory) -> physical GPU (device: 0, name: Tesla K80, pci bus id: 0000:00:04.0, compute capability: 3.7)\n",
            "W0906 11:40:00.697411 140289655293824 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/saver.py:1276: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use standard file APIs to check for files with this prefix.\n",
            "EPOCH 1 ...\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "2019-09-06 11:40:12.727689: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudnn.so.7\n",
            "2019-09-06 11:40:18.596945: W tensorflow/core/common_runtime/bfc_allocator.cc:237] Allocator (GPU_0_bfc) ran out of memory trying to allocate 1.33GiB with freed_by_count=0. The caller indicates that this is not a failure, but may mean that there could be performance gains if more memory were available.\n",
            "Loss: = 0.094\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.059\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.100\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.054\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.071\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.038\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.072\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.056\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.059\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.064\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.069\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.066\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n",
            "Loss: = 0.070\n",
            "(4, 704, 1280, 3) (4, 704, 1280, 2)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zd-iCXO5Aqs7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nQzfvytAAqqk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KwlQ3ClDAqn2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CHb83_i6Aqk7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V4lxGvG169o8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import os\n",
        "import numpy as np\n",
        "from PIL import Image"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Q1QHVRZhPCUK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from matplotlib.pyplot import imshow\n",
        "%matplotlib inline"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5T-KqaGIpSmc",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "import re\n",
        "import random\n",
        "from glob import glob\n",
        "import shutil\n",
        "import time"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "REB561v21Obe",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import tensorflow as tf"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-okVPNx5rJIc",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "BACKGROUND_COLOR = [255, 0, 0]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FaoYafIzqiOx",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def one_hot(gt_image):\n",
        "  gt_bg = np.all(gt_image == BACKGROUND_COLOR, axis=2)\n",
        "  gt_bg = gt_bg.reshape(*gt_bg.shape, 1)\n",
        "  gt_image = np.concatenate((gt_bg, np.invert(gt_bg)), axis=2)\n",
        "  \n",
        "  return gt_image"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MFpxRYa6k2OD",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def preprocess_image(image, gt_image, width, height):\n",
        "  \n",
        "  image = image.resize((width, height), Image.ANTIALIAS)    # best down-sizing filter\n",
        "  gt_image = gt_image.resize((width, height), Image.ANTIALIAS)    # best down-sizing filter\n",
        "  \n",
        "  image = np.float32(image)\n",
        "  image = image/255.0\n",
        "  \n",
        "  gt_image = np.array(gt_image)\n",
        "  gt_image = np.float32(one_hot(gt_image))\n",
        "  \n",
        "  return image, gt_image\n",
        "  "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1rVo4WI3JrI4",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def gen_batch_fn(data_dir, width = 1200, height = 350):\n",
        "  \n",
        "  image_paths = glob(os.path.join(data_dir, 'image_2', '*.png'))        \n",
        "  gt_paths = {\n",
        "      re.sub(r'_(lane|road)_', '_', os.path.basename(path)): path \n",
        "      for path in glob(os.path.join(data_dir, 'gt_image_2', '*_road_*.png'))\n",
        "  }\n",
        "  \n",
        "  background_color = np.array([255, 0, 0])\n",
        "  \n",
        "  def get_batch(batch_size):\n",
        "    \n",
        "    random.shuffle(image_paths)\n",
        "    for batch in range(0, len(image_paths), batch_size):\n",
        "      images = []\n",
        "      gt_images = []\n",
        "      \n",
        "      for image_file in image_paths[batch: batch + batch_size]:\n",
        "        gt_file = gt_paths[os.path.basename(image_file)]\n",
        "        \n",
        "        image = Image.open(image_file)\n",
        "        gt_image = Image.open(gt_file)\n",
        "        \n",
        "        image, gt_image = preprocess_image(image, gt_image, width, height)\n",
        "        \n",
        "        #print(image.shape, gt_image.shape)\n",
        "        \n",
        "        images.append(image)\n",
        "        gt_images.append(gt_image)\n",
        "        \n",
        "      images = np.array(images)\n",
        "      gt_images = np.array(gt_images) \n",
        "      \n",
        "      yield images, gt_images\n",
        "        \n",
        "  \n",
        "  return get_batch"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZE5tcNuqE2Ft",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# image_paths = glob(os.path.join('data/potholes/idd', '*.png'))\n",
        "# paths = []\n",
        "# for img_path in image_paths:\n",
        "#   m = os.path.basename(img_path).split('_')[0]\n",
        "#   st = m + '_gtFine_polygons.png'\n",
        "#   x = os.path.join('data/IDD/gtFine/train/*', st)\n",
        "#   s = glob(x)\n",
        "#   [g] = s\n",
        "#   print(g)\n",
        "#   break"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G03-DKiADuTb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def gen_batch_fn_tempdemo(data_dir, width = 1200, height = 350, datatype='train'):\n",
        "  \n",
        "  background_color = np.array([255, 0, 0])\n",
        "  image_paths = glob(os.path.join('data/potholes/australia', '*.JPG'))\n",
        "  paths = []\n",
        "  for img_path in image_paths:\n",
        "    m = os.path.basename(img_path).split('_')[0]\n",
        "    st = m + '_gtFine_polygons.png'\n",
        "    x = os.path.join('data/IDD/gtFine/train/*', st)\n",
        "    s = glob(x)\n",
        "    [gt_path] = s \n",
        "    paths.append([img_path, gt_path])\n",
        "  \n",
        "  def get_batch(batch_size):\n",
        "    \n",
        "    random.shuffle(paths)\n",
        "    for batch in range(0, len(paths), batch_size):\n",
        "      images = []\n",
        "      gt_images = []\n",
        "      \n",
        "      for path in paths[batch: batch + batch_size]:\n",
        "        \n",
        "        image = Image.open(path[0])\n",
        "        gt_image = Image.open(path[1])\n",
        "        \n",
        "        image, gt_image = preprocess_image(image, gt_image, width, height)\n",
        "                \n",
        "        images.append(image)\n",
        "        gt_images.append(gt_image)\n",
        "        \n",
        "      images = np.array(images)\n",
        "      gt_images = np.array(gt_images) \n",
        "      \n",
        "      yield images, gt_images\n",
        "  \n",
        "  return get_batch"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WK3WU5V3BCP8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def gen_batch_fn_idd(data_dir, width = 1200, height = 350, datatype='train'):\n",
        "  \n",
        "  background_color = np.array([255, 0, 0])\n",
        "  image_paths = glob(os.path.join(data_dir, 'leftImg8bit', datatype, '*', '*.png'))\n",
        "  paths = []\n",
        "  for img_path in image_paths:\n",
        "    m = os.path.basename(img_path).split('_')[0]\n",
        "    st = m + '_gtFine_polygons.png'\n",
        "    n = os.path.dirname(img_path).split('/')\n",
        "    n[2] = 'gtFine'\n",
        "    n = '/'.join(n)\n",
        "    gt_path = os.path.join(n, st)\n",
        "    paths.append([img_path, gt_path])\n",
        "  \n",
        "  def get_batch(batch_size):\n",
        "    \n",
        "    random.shuffle(paths)\n",
        "    for batch in range(0, len(paths), batch_size):\n",
        "      images = []\n",
        "      gt_images = []\n",
        "      \n",
        "      for path in paths[batch: batch + batch_size]:\n",
        "        \n",
        "        image = Image.open(path[0])\n",
        "        gt_image = Image.open(path[1])\n",
        "        \n",
        "        image, gt_image = preprocess_image(image, gt_image, width, height)\n",
        "                \n",
        "        images.append(image)\n",
        "        gt_images.append(gt_image)\n",
        "        \n",
        "      images = np.array(images)\n",
        "      gt_images = np.array(gt_images) \n",
        "      \n",
        "      yield images, gt_images\n",
        "  \n",
        "  return get_batch\n",
        "    "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pXsJe90BGI5q",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# get_batch = gen_batch_fn_idd(data_dir='data/IDD', width = 1200, height = 350, datatype='train')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vOQmjPDotHDK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# get_batch = gen_batch_fn('data/data_road/training', width = 1120, height = 256)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Jpyfv8VpthSA",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# for ims, gt_ims in get_batch(2):\n",
        "#   print(ims.shape, gt_ims.shape)\n",
        "#   #print(gt_ims)\n",
        "  \n",
        "#   break"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3Cxb4jVZzcXj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def save_test_predictions(sess, test_dir, save_dir, width = 1120, height = 256):\n",
        "  output_dir = os.path.join(save_dir, str(time.time()))\n",
        "  if os.path.exists(output_dir):\n",
        "    shutil.rmtree(output_dir)\n",
        "  os.makedirs(output_dir)\n",
        "  \n",
        "  print(\"Training finished. Saving images to {}\".format(output_dir))\n",
        "  \n",
        "  for image in glob(os.path.join(test_dir, 'image_2', '*.png')):\n",
        "    im = Image.open(image)\n",
        "    im = im.resize((width, height), Image.ANTIALIAS)\n",
        "    im_arr = np.array(im)\n",
        "    \n",
        "    im_scaled = np.float32(im_arr)/255.0\n",
        "    \n",
        "    label = sess.run([tf.math.softmax(output_label)], feed_dict = {input_im: [im_scaled]})\n",
        "    print(len(label))\n",
        "    label = label[0][:, :, 1]\n",
        "    segmentation = (label > 0.5).reshape(height, width, 1)\n",
        "    \n",
        "    mask = np.dot(segmentation, np.array([[0, 255, 0, 127]]))    \n",
        "    mask = Image.fromarray(mask, 'RGBA')\n",
        "    \n",
        "    Image.paste(im, mask = mask)\n",
        "    \n",
        "    im.save(os.path.join(save_dir, os.path.basename(image)))\n",
        "    \n",
        "    break\n",
        "    "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Fxd1Nke0_S6u",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bIGoQHLP0eFh",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "from tensorflow.contrib import slim\n",
        "from tensorflow.contrib.slim.nets import resnet_v2\n",
        "#from models import resnet_v2"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QLs3kSUuFf4Q",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from tensorflow.keras import layers"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4kIc4Evb_MXw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oFlkeE8USbxR",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "tf.reset_default_graph()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_Wa9TOVM0dpQ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VebHnbdoDwhP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def arm_module(layer, n_filter_maps):\n",
        "  \n",
        "  net = tf.math.reduce_mean(layer, axis = [1,2], keepdims = True, name = 'global_avg_pool')\n",
        "  net = layers.Conv2D(filters = n_filter_maps, kernel_size = [1,1])(net)\n",
        "  net = layers.BatchNormalization()(net)\n",
        "  net = tf.math.sigmoid(net, name = 'sigmoid')\n",
        "  \n",
        "  scaled_layer = tf.multiply(layer, net)\n",
        "  \n",
        "  return scaled_layer"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-FTzVxWRJW9T",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def ffm_module(spatial_layer, context_layer, num_classes):\n",
        "  \n",
        "  input_features = tf.concat([spatial_layer, context_layer], axis = -1)\n",
        "  net = layers.Conv2D(filters = num_classes, kernel_size = [3,3], padding = 'same', activation = 'relu')(input_features)\n",
        "  net = layers.BatchNormalization()(net)\n",
        "  \n",
        "  net_vector = tf.reduce_mean(net, axis = [1,2], keepdims = True)\n",
        "  \n",
        "  # First 1x1 convolution uses 16 filters\n",
        "  net_vector = layers.Conv2D(filters = 16, kernel_size = [1,1], padding = 'same', activation = 'relu')(net_vector)\n",
        "  net_vector = layers.Conv2D(filters = num_classes, kernel_size = [1,1], padding = 'same')(net_vector)\n",
        "  net_vector = layers.Activation('sigmoid')(net_vector)\n",
        "  \n",
        "  net_scaled = tf.multiply(net, net_vector)\n",
        "  net = tf.add(net, net_scaled)\n",
        "  \n",
        "  return net\n",
        "  "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wzPe9OzB7RsL",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "frontend = 'ResNet101'\n",
        "num_classes = 2\n",
        "input_im = tf.placeholder(shape = (None, None, None, 3), dtype = tf.float32, name='input_im')\n",
        "gt_im = tf.placeholder(shape = (None, None, None, 2), dtype = tf.float32, name='gt_im')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dDARlPErrT07",
        "colab_type": "code",
        "outputId": "c427492a-09b8-4d1b-dbc5-2c20b2c9a680",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(input_im)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"input_im:0\", shape=(?, ?, ?, 3), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hjDYIGruGRT7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "## Spatial Path"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HTsnXot6GS2o",
        "colab_type": "code",
        "outputId": "b7550883-762b-4ffe-fb41-aa1fda336c7a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 108
        }
      },
      "source": [
        "layer_spatial = layers.Conv2D(input_shape=(None, 1120, 256, 3), filters=64, kernel_size=[3,3], strides=2, padding='same', activation='relu')(input_im)\n",
        "layer_spatial = layers.Conv2D(filters=128, kernel_size=[3,3], strides=2, padding='same', activation='relu')(layer_spatial)\n",
        "spatial_output = layers.Conv2D(filters=256, kernel_size=[3,3], strides=2, padding='same', activation='relu')(layer_spatial)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING: Logging before flag parsing goes to stderr.\n",
            "W0831 08:49:35.604157 140356339836800 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:1251: calling VarianceScaling.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kMV_eeBFIvhE",
        "colab_type": "code",
        "outputId": "86ed1309-8055-44c0-cb6f-a3b3a8fea907",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "spatial_output"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Tensor 'conv2d_2/Relu:0' shape=(?, ?, ?, 256) dtype=float32>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xwqnJpqJGR5v",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "## Context Path"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gzHJSL5Ajvzw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "with slim.arg_scope(resnet_v2.resnet_arg_scope()):\n",
        "  #tf.reset_default_graph()\n",
        "  last_layer, end_points = resnet_v2.resnet_v2_101(input_im, is_training=True, scope='resnet_v2_101', global_pool = False)\n",
        "  frontend_scope='resnet_v2_101'\n",
        "  init_fn = slim.assign_from_checkpoint_fn(model_path=os.path.join('models', 'resnet_v2_101.ckpt'), var_list=slim.get_model_variables('resnet_v2_101'), ignore_missing_vars=True)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Zj1y6ofJ8IwP",
        "colab_type": "code",
        "outputId": "bc9e8ef6-20b7-4eea-b084-8b468a6f91d0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(last_layer)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/postnorm/Relu:0\", shape=(?, ?, ?, 2048), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9B7t05BN_ifa",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "layer_reduced16 = end_points[frontend_scope + '/block2']\n",
        "layer_reduced32 = last_layer"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "F09EOrePBDkY",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "layer_arm16 = arm_module(layer_reduced16, n_filter_maps = 512)\n",
        "layer_arm32 = arm_module(layer_reduced32, n_filter_maps = 2048)\n",
        "layer_global_context = tf.reduce_mean(last_layer, axis = [1,2], keepdims = True, name = 'global_context')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yKvfy4yXy8CD",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "## Combining Context Features\n",
        "layer_context1 = tf.math.multiply(layer_arm32, layer_global_context)\n",
        "layer_context1 = layers.UpSampling2D(size = 4, interpolation = 'bilinear')(layer_context1)\n",
        "layer_context2 = layers.UpSampling2D(size = 2, interpolation = 'bilinear')(layer_arm16)\n",
        "\n",
        "context_output = tf.concat([layer_context1, layer_context2], axis = -1)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "q3OC2ifOFepl",
        "colab_type": "code",
        "outputId": "60b62762-be8a-436d-90db-3017b796e00c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(context_output)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"concat:0\", shape=(?, ?, ?, 2560), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "34RZteHdE9rB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "ffm_output = ffm_module(spatial_output, context_output, num_classes)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n1ULQeqbJU2D",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "output_label = layers.UpSampling2D(size = 8, interpolation = 'bilinear')(ffm_output)\n",
        "output_label = layers.Conv2D(filters = num_classes, kernel_size = [1,1], activation = None)(output_label)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wI95ffITJQtj",
        "colab_type": "code",
        "outputId": "db9ab478-4db3-49a0-ccce-5a4ecc36e32e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(output_label)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"conv2d_8/BiasAdd:0\", shape=(?, ?, ?, 2), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "f4EenkOSUBG2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "pred_gt = tf.math.softmax(output_label)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RoifS9uQVykC",
        "colab_type": "code",
        "outputId": "8df0e931-83ec-41f6-e421-cfdb59700f0d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "pred_gt"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Tensor 'Softmax:0' shape=(?, ?, ?, 2) dtype=float32>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 121
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z3rngRvWUpZW",
        "colab_type": "code",
        "outputId": "9370f9e9-a23d-4120-bd7b-d334cbd18e6d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "print(pred_gt)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"Softmax:0\", shape=(?, ?, ?, 2), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7TiVSXe7A9Ge",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "EPOCHS = 10\n",
        "BATCH_SIZE = 8\n",
        "TEST_DIR = 'data/data_road/testing'\n",
        "SAVE_DIR = 'saved_tests'"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jskeFke0WQ5F",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "EPOCHS = 10\n",
        "BATCH_SIZE = 4"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9tm2NvMTA84d",
        "colab_type": "code",
        "outputId": "94a40d7f-7769-4f0c-f30c-a1b183f22405",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 143
        }
      },
      "source": [
        "cross_entropy_loss = tf.reduce_mean(tf.losses.softmax_cross_entropy(onehot_labels = gt_im, logits = output_label))\n",
        "# optimizer = tf.train.AdamOptimizer(learning_rate = 0.001).minimize(cross_entropy_loss)\n",
        "optimizer = tf.train.RMSPropOptimizer(learning_rate = 0.0001, decay = 0.995).minimize(cross_entropy_loss)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "W0831 08:49:44.788798 140356339836800 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/losses/losses_impl.py:121: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
            "W0831 08:49:47.361911 140356339836800 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/rmsprop.py:119: calling Ones.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2B8a_k5Gf4Gd",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "saver = tf.train.Saver()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uYtANDf1Bx2E",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "sess = tf.Session()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "q18sElYf3aq_",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)\n",
        "run_metadata = tf.RunMetadata()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1G1Ktoh8OyHm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#with tf.Session() as sess:\n",
        "sess.run(tf.global_variables_initializer())\n",
        "init_fn(sess)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mT5ZyzL4soVY",
        "colab_type": "code",
        "outputId": "873f6f21-1230-483c-e560-cd09c411f1c0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 90
        }
      },
      "source": [
        "saver.restore(sess, \"./bisenet_ckpt/model\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "W0803 03:36:06.193031 139652218144640 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/saver.py:1276: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use standard file APIs to check for files with this prefix.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "azOI_ECx2o1c",
        "colab_type": "code",
        "outputId": "848155e4-8a40-40cc-a3b6-9050bfc2767e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 90
        }
      },
      "source": [
        "saver.restore(sess, \"./bisenet_ckpt/model_idd2\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "W0831 08:49:55.271504 140356339836800 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/saver.py:1276: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use standard file APIs to check for files with this prefix.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lFfoQbWqOgBi",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "saver.restore(sess, \"./bisenet_ckpt/model_idd\")"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LmmBeKqcMqsj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "save_path = saver.save(sess, \"./bisenet_ckpt/model_idd2\")"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "heDlnjcV4ivf",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from tensorflow.python.client import timeline"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5pYnX3XRA8sB",
        "colab_type": "code",
        "outputId": "22cad5e0-0cdb-4f53-f213-943407021295",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 848
        }
      },
      "source": [
        "get_batch = gen_batch_fn_idd('data/IDD', width = 1280, height = 704)\n",
        "j = 0\n",
        "for i in range(EPOCHS):\n",
        "  print(\"EPOCH {} ...\".format(i+1))\n",
        "  for image, label in get_batch(BATCH_SIZE):\n",
        "      s = time.time()\n",
        "      _, loss = sess.run([optimizer, cross_entropy_loss], \n",
        "                         feed_dict={input_im: image, gt_im: label})\n",
        "#                         options=run_options, run_metadata=run_metadata)\n",
        "      print(\"Loss: = {:.3f}\".format(loss))\n",
        "      print(time.time() - s)\n",
        "      \n",
        "#       if j==5:\n",
        "#         tl = timeline.Timeline(run_metadata.step_stats)\n",
        "#         ctf = tl.generate_chrome_trace_format()\n",
        "#         with open('timeline.json', 'w') as f:\n",
        "#             f.write(ctf)\n",
        "#         print(\"Done\")\n",
        "#         break\n",
        "#       j+=1\n",
        "#   print()\n",
        "#   break\n",
        "\n",
        "#   if i%10 == 0:\n",
        "  save_path = saver.save(sess, \"./bisenet_ckpt/model_idd2\")\n",
        "  #save_test_predictions(sess, TEST_DIR, SAVE_DIR, width = 1120, height = 256)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "EPOCH 1 ...\n",
            "Loss: = 0.067\n",
            "6.860572814941406\n",
            "Loss: = 0.134\n",
            "1.0919506549835205\n",
            "Loss: = 0.219\n",
            "1.0884120464324951\n",
            "Loss: = 0.203\n",
            "1.091390609741211\n",
            "Loss: = 0.065\n",
            "1.0887327194213867\n",
            "Loss: = 0.098\n",
            "1.091291904449463\n",
            "Loss: = 0.061\n",
            "1.088341474533081\n",
            "Loss: = 0.414\n",
            "1.095041036605835\n",
            "Loss: = 0.168\n",
            "1.0904781818389893\n",
            "Loss: = 0.106\n",
            "1.0923280715942383\n",
            "Loss: = 0.181\n",
            "1.0934407711029053\n",
            "Loss: = 0.096\n",
            "1.0874485969543457\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-58-01136b5d963e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      6\u001b[0m       \u001b[0ms\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtime\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtime\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      7\u001b[0m       _, loss = sess.run([optimizer, cross_entropy_loss], \n\u001b[0;32m----> 8\u001b[0;31m                          feed_dict={input_im: image, gt_im: label})\n\u001b[0m\u001b[1;32m      9\u001b[0m \u001b[0;31m#                         options=run_options, run_metadata=run_metadata)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     10\u001b[0m       \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"Loss: = {:.3f}\"\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mformat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloss\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36mrun\u001b[0;34m(self, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m    948\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    949\u001b[0m       result = self._run(None, fetches, feed_dict, options_ptr,\n\u001b[0;32m--> 950\u001b[0;31m                          run_metadata_ptr)\n\u001b[0m\u001b[1;32m    951\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    952\u001b[0m         \u001b[0mproto_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_run\u001b[0;34m(self, handle, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m   1171\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mfinal_fetches\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0mfinal_targets\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mhandle\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mfeed_dict_tensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1172\u001b[0m       results = self._do_run(handle, final_targets, final_fetches,\n\u001b[0;32m-> 1173\u001b[0;31m                              feed_dict_tensor, options, run_metadata)\n\u001b[0m\u001b[1;32m   1174\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1175\u001b[0m       \u001b[0mresults\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_do_run\u001b[0;34m(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m   1348\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mhandle\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1349\u001b[0m       return self._do_call(_run_fn, feeds, fetches, targets, options,\n\u001b[0;32m-> 1350\u001b[0;31m                            run_metadata)\n\u001b[0m\u001b[1;32m   1351\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1352\u001b[0m       \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_do_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0m_prun_fn\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeeds\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetches\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_do_call\u001b[0;34m(self, fn, *args)\u001b[0m\n\u001b[1;32m   1354\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_do_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1355\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1356\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1357\u001b[0m     \u001b[0;32mexcept\u001b[0m \u001b[0merrors\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mOpError\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1358\u001b[0m       \u001b[0mmessage\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcompat\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mas_text\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmessage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_run_fn\u001b[0;34m(feed_dict, fetch_list, target_list, options, run_metadata)\u001b[0m\n\u001b[1;32m   1339\u001b[0m       \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_extend_graph\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1340\u001b[0m       return self._call_tf_sessionrun(\n\u001b[0;32m-> 1341\u001b[0;31m           options, feed_dict, fetch_list, target_list, run_metadata)\n\u001b[0m\u001b[1;32m   1342\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1343\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_prun_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_call_tf_sessionrun\u001b[0;34m(self, options, feed_dict, fetch_list, target_list, run_metadata)\u001b[0m\n\u001b[1;32m   1427\u001b[0m     return tf_session.TF_SessionRun_wrapper(\n\u001b[1;32m   1428\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_session\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moptions\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtarget_list\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1429\u001b[0;31m         run_metadata)\n\u001b[0m\u001b[1;32m   1430\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1431\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_call_tf_sessionprun\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lIgMzfCFfUWT",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "sess.close()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sMN0NMIIBWzF",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "save_dir = SAVE_DIR\n",
        "test_dir = TEST_DIR\n",
        "width = 1280\n",
        "height = 704"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "flK_KJmxSAlM",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "im_names1 = []\n",
        "res = ['351875_leftImg8bit.png', '380341_leftImg8bit.png', '416335_leftImg8bit.png', '456726_leftImg8bit.png', '465708_leftImg8bit.png']\n",
        "x = glob('/content/gdrive/My Drive/SemanticSegmentation/data/IDD/leftImg8bit/train/*/*.png')\n",
        "for i in x:\n",
        "  if os.path.basename(i) in res:\n",
        "    im_names1.append(i)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "s7LjY6VsSCy7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "x = glob(os.path.join('/content/gdrive/My Drive/SemanticSegmentation/data/IDD/leftImg8bit/val/*', '*.png'))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "D1buQCuAC8rM",
        "colab_type": "code",
        "outputId": "c3d926d8-5f49-4f40-d9f8-1bf6d1918cb3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "\n",
        "## Changes on Aug 18\n",
        "# mask_dir = os.path.join(save_dir, 'final_results')\n",
        "# if os.path.exists(mask_dir):\n",
        "#   shutil.rmtree(mask_dir)\n",
        "# os.makedirs(mask_dir)\n",
        "##\n",
        "output_dir = os.path.join(save_dir, 'final_results_sa')\n",
        "if os.path.exists(output_dir):\n",
        "  shutil.rmtree(output_dir)\n",
        "os.makedirs(output_dir)\n",
        "\n",
        "\n",
        "print(\"Saving images to {}\".format(output_dir))\n",
        "\n",
        "# start_time = time.time()\n",
        "\n",
        "i = 0\n",
        "# start_time = time.time()\n",
        "\n",
        "\n",
        "# for image in im_names1:\n",
        "for image in glob(os.path.join('/content/gdrive/My Drive/Pothole-Detection-Retina/keras-retinanet/data/test', '*.JPG')):\n",
        "  im = Image.open(image)\n",
        "  im = im.resize((width, height), Image.ANTIALIAS)\n",
        "\n",
        "  im_scaled = np.float32(im)/255.0\n",
        "  \n",
        "  inputs = np.array([im_scaled])\n",
        "  \n",
        "  \n",
        "  \n",
        "  label = sess.run(tf.math.softmax(output_label), feed_dict = {input_im: inputs},)\n",
        "#   print(time.time() - start_time)\n",
        "#   test_label = label\n",
        "  #print(np.array(label).shape)\n",
        "#   label = label[0][:, :, 1]\n",
        "#   segmentation = (label > 0.5).reshape(height, width, 1)\n",
        "  \n",
        "  segmentation = improve_crf(inputs, label)\n",
        "\n",
        "  mask = np.dot(segmentation, np.array([[0, 255, 0, 60]]))\n",
        "  mask = mask.astype('uint8')\n",
        "  mask = Image.fromarray(mask, 'RGBA')\n",
        "\n",
        "  # MODIFIED 3 AUG\n",
        "  im.paste(mask, mask = mask)\n",
        "\n",
        "  im_name = os.path.basename(image)\n",
        "  im_arr = im_name.split('.')\n",
        "  mask_name = im_arr[0] + '_mask.png'\n",
        "\n",
        "  mask.save(os.path.join(output_dir, mask_name))\n",
        "  im.save(os.path.join(output_dir, os.path.basename(image)))\n",
        "  \n",
        "print(\"Saved\")  \n",
        "#   print(time.time() - start_time)\n",
        "  \n",
        "#   start_time = time.time()\n",
        "\n",
        "  \n",
        "#   tl = timeline.Timeline(run_metadata.step_stats)\n",
        "#   ctf = tl.generate_chrome_trace_format()\n",
        "#   m = 'timelineg' + str(i) + '.json'\n",
        "#   with open(m, 'w') as f:\n",
        "#       f.write(ctf)\n",
        "#   i+=1\n",
        "#   print(\"--- %s seconds ---\" % (time.time() - start_time))\n",
        "  "
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Saving images to saved_tests/final_results_sa\n",
            "Saved\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AS0WntTtcdcf",
        "colab_type": "code",
        "outputId": "a649d859-7c70-42fa-e795-f2e224d9bf88",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "x = glob('/content/gdrive/My Drive/SemanticSegmentation/data/IDD/leftImg8bit/val/*/401205_leftImg8bit.png')\n",
        "x"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['/content/gdrive/My Drive/SemanticSegmentation/data/IDD/leftImg8bit/val/21/401205_leftImg8bit.png']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 72
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K3FyHswvac_f",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "output_dir = os.path.join(save_dir, 'testing')\n",
        "# if os.path.exists(output_dir):\n",
        "#   shutil.rmtree(output_dir)\n",
        "# os.makedirs(output_dir)\n",
        "\n",
        "\n",
        "im = Image.open(x[0])\n",
        "im = im.resize((width, height), Image.ANTIALIAS)\n",
        "\n",
        "im_scaled = np.float32(im)/255.0\n",
        "\n",
        "inputs = np.array([im_scaled])\n",
        "\n",
        "\n",
        "\n",
        "label = sess.run(tf.math.softmax(output_label), feed_dict = {input_im: inputs},)\n",
        "#   print(time.time() - start_time)\n",
        "#   test_label = label\n",
        "#print(np.array(label).shape)\n",
        "# label = label[0][:, :, 1]\n",
        "# segmentation = (label > 0.5).reshape(height, width, 1)\n",
        "\n",
        "\n",
        "segmentation = improve_crf(inputs, label)\n",
        "# background = 1 - road\n",
        "# road = np.dot(road, np.array([[255, 0, 255]]))\n",
        "# background = np.dot(background, np.array([[255, 0, 0]]))\n",
        "\n",
        "mask = np.dot(segmentation, np.array([[0, 255, 0, 60]]))\n",
        "mask = mask.astype('uint8')\n",
        "mask = Image.fromarray(mask, 'RGBA')\n",
        "\n",
        "# # MODIFIED 3 AUG\n",
        "im.paste(mask, mask = mask)\n",
        "\n",
        "# im_name = os.path.basename(image)\n",
        "# im_arr = im_name.split('.')\n",
        "# mask_name = im_arr[0] + '_mask.png'\n",
        "\n",
        "# #   mask.save(os.path.join(output_dir, mask_name))\n",
        "im.save(os.path.join(output_dir, os.path.basename(x[0])))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YRFa0hh0eE4n",
        "colab_type": "code",
        "outputId": "11429b32-6983-4304-8e2a-cfe4ccbd0d75",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "label.shape"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1, 704, 1280, 2)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 63
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c1Wv3HWgcvMq",
        "colab_type": "code",
        "outputId": "aa610bca-f219-4604-8560-be7d6a5a0f84",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "segmentation.shape"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(704, 1280, 1)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 51
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dcARKddwzZQn",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-iaLDt4VfHel",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "height = 704\n",
        "width = 1280\n",
        "iou_arr = []\n",
        "get_batch = gen_batch_fn_idd('data/IDD', width = width, height = height, datatype='val')\n",
        "for image, label in get_batch(1):\n",
        "  prediction = sess.run(tf.math.softmax(output_label), feed_dict = {input_im: image})\n",
        "#   road = prediction[0][:, :, 1]\n",
        "#   background = prediction[0][:, :, 0]\n",
        "  road_target = label[0][:, :, 1].reshape(height, width, 1)\n",
        "  background_target = label[0][:, :, 0].reshape(height, width, 1)\n",
        "  \n",
        "#   road = (road > 0.5).reshape(height, width, 1)\n",
        "#   background = (background > 0.5).reshape(height, width, 1)\n",
        "  \n",
        "#   road = np.dot(road, np.array([[255, 0, 255]]))\n",
        "#   background = np.dot(background, np.array([[255, 0, 0]]))\n",
        "  road_target = np.dot(road_target, np.array([[255, 0, 255]]))\n",
        "  background_target = np.dot(background_target, np.array([[255, 0, 0]]))\n",
        "  \n",
        "  road = improve_crf(image, prediction)\n",
        "  background = 1 - road\n",
        "  road = np.dot(road, np.array([[255, 0, 255]]))\n",
        "  background = np.dot(background, np.array([[255, 0, 0]]))\n",
        "#   pred_im = x + y\n",
        "  \n",
        "  pred_im = road + background\n",
        "  target_im = road_target + background_target\n",
        "  \n",
        "    \n",
        "  intersection = np.logical_and(target_im, pred_im)\n",
        "  union = np.logical_or(target_im, pred_im)\n",
        "  iou_score = np.sum(intersection) / np.sum(union)\n",
        "  \n",
        "  iou_arr.append(iou_score)\n",
        "\n",
        "  print(iou_score)\n",
        "  \n",
        "  ## 28-08\n",
        "#   break"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J_3yAEwj6Zda",
        "colab_type": "code",
        "outputId": "a5926ce3-8dc7-4e22-af52-91c08a9668c3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "len(iou_arr)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "981"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 112
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G0F77fUu4D_t",
        "colab_type": "code",
        "outputId": "26a2aaf4-d249-4783-f4a6-9e4d85b6c646",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "## With CRF\n",
        "np.mean(iou_arr)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.9626840999721897"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 113
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DqvzG9Q4GVWn",
        "colab_type": "code",
        "outputId": "b3a01650-183d-47ba-aa12-3e5c03bb8825",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "## Without CRF\n",
        "\n",
        "np.mean(iou_arr)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.9654240350647626"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 46
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lv-obqjcB4t4",
        "colab_type": "code",
        "outputId": "16cb5739-2104-44ee-81c2-934a925384d5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "!ls"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "bisenet_ckpt  crf  data  models  runs  runs2  saved_tests\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VqRu__RiB3cj",
        "colab_type": "code",
        "outputId": "65c07f01-f885-4be4-dbaf-abaae4b4b48f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 214
        }
      },
      "source": [
        "pip install git+https://github.com/lucasb-eyer/pydensecrf.git"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting git+https://github.com/lucasb-eyer/pydensecrf.git\n",
            "  Cloning https://github.com/lucasb-eyer/pydensecrf.git to /tmp/pip-req-build-_8866yt4\n",
            "  Running command git clone -q https://github.com/lucasb-eyer/pydensecrf.git /tmp/pip-req-build-_8866yt4\n",
            "Building wheels for collected packages: pydensecrf\n",
            "  Building wheel for pydensecrf (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for pydensecrf: filename=pydensecrf-1.0rc2-cp36-cp36m-linux_x86_64.whl size=2235008 sha256=ebcd0e24f02aeb72091f90e408fd43e5ea31a9474b613cee820e0d01685e3d9e\n",
            "  Stored in directory: /tmp/pip-ephem-wheel-cache-rcdpml2s/wheels/81/ac/ed/09f1890b954ca1ff7f189518cc9a319138e50e7ea88937fae9\n",
            "Successfully built pydensecrf\n",
            "Installing collected packages: pydensecrf\n",
            "Successfully installed pydensecrf-1.0rc2\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MDcoBELldZlU",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import sys\n",
        "\n",
        "# path = \"/content/gdrive/My Drive/SemanticSegmentation/crf/pydensecrf/\"\n",
        "# sys.path.append(path)\n",
        "\n",
        "import pydensecrf.densecrf as dcrf\n",
        "\n",
        "from pydensecrf.utils import compute_unary, create_pairwise_bilateral, \\\n",
        "    create_pairwise_gaussian, unary_from_softmax\n",
        "\n",
        "import skimage.io as io\n",
        "\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_Et9-hIG_FMU",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def improve_crf(image, prediction):\n",
        "  image_crf = image[0]\n",
        "\n",
        "  softmax = prediction[0].squeeze()\n",
        "\n",
        "  softmax = softmax.transpose((2, 0, 1))\n",
        "\n",
        "  # The input should be the negative of the logarithm of probability values\n",
        "  # Look up the definition of the softmax_to_unary for more information\n",
        "  unary = unary_from_softmax(softmax)\n",
        "\n",
        "  # The inputs should be C-continious -- we are using Cython wrapper\n",
        "  unary = np.ascontiguousarray(unary)\n",
        "\n",
        "  d = dcrf.DenseCRF(image_crf.shape[0] * image_crf.shape[1], 2)\n",
        "\n",
        "  d.setUnaryEnergy(unary)\n",
        "\n",
        "  # This potential penalizes small pieces of segmentation that are\n",
        "  # spatially isolated -- enforces more spatially consistent segmentations\n",
        "  feats = create_pairwise_gaussian(sdims=(10, 10), shape=image_crf.shape[:2])\n",
        "\n",
        "  d.addPairwiseEnergy(feats, compat=3,\n",
        "                      kernel=dcrf.DIAG_KERNEL,\n",
        "                      normalization=dcrf.NORMALIZE_SYMMETRIC)\n",
        "\n",
        "  # This creates the color-dependent features --\n",
        "  # because the segmentation that we get from CNN are too coarse\n",
        "  # and we can use local color features to refine them\n",
        "  feats = create_pairwise_bilateral(sdims=(50, 50), schan=(20, 20, 20),\n",
        "                                     img=image_crf, chdim=2)\n",
        "\n",
        "  d.addPairwiseEnergy(feats, compat=10,\n",
        "                       kernel=dcrf.DIAG_KERNEL,\n",
        "                       normalization=dcrf.NORMALIZE_SYMMETRIC)\n",
        "  Q = d.inference(5)\n",
        "\n",
        "  res = np.argmax(Q, axis=0).reshape((image_crf.shape[0], image_crf.shape[1]))\n",
        "  \n",
        "  return res.reshape((image_crf.shape[0], image_crf.shape[1], 1))\n",
        "\n",
        "  # cmap = plt.get_cmap('bwr')\n",
        "\n",
        "  # f, (ax1, ax2) = plt.subplots(1, 2, sharey=True)\n",
        "  # ax1.imshow(res, vmax=1.5, vmin=-0.4, cmap=cmap)\n",
        "  # ax1.set_title('Segmentation with CRF post-processing')\n",
        "  # probability_graph = ax2.imshow(np.dstack((label[0],)*3)*100)\n",
        "  # ax2.set_title('Ground-Truth Annotation')\n",
        "  # plt.show()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "39rptkXdwrWM",
        "colab_type": "code",
        "outputId": "92d7a1c6-bb31-4921-f3ec-c6e0af7b42ad",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "res = res.reshape(image_crf.shape[0], image_crf.shape[1], 1)\n",
        "print(res.shape)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(704, 1280, 1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BTjAaIRkxAMz",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YzRfDpplxl1Q",
        "colab_type": "code",
        "outputId": "701708a8-1e16-471d-f35b-e1065d4d84cd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 252
        }
      },
      "source": [
        "imshow(pred_im)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f28010a0390>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 102
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAADbCAYAAABjl1pvAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0vUEw-ZxJVK-",
        "colab_type": "code",
        "outputId": "fe8abd81-1bf7-475f-fe72-866d5552b872",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "res.reshape((image_crf.shape[0], image_crf.shape[1], 1)\n",
        "road = np.dot(road, np.array([[255, 0, 255]]))\n",
        "background = np.dot(background, np.array([[255, 0, 0]]))\n",
        "road_target = np.dot(road_target, np.array([[255, 0, 255]]))\n",
        "background_target = np.dot(background_target, np.array([[255, 0, 0]]))\n",
        "\n",
        "pred_im = road + background"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(704, 1280)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 64
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8bc2rKqJqHmO",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "sess.close()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8jZfk7gw_FKl",
        "colab_type": "code",
        "outputId": "4401297f-3895-43de-e25a-63b40227fc1b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "label[0].shape"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(704, 1280, 2)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 70
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wTzyZC-r_FHi",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W4TUUvqJC8h5",
        "colab_type": "code",
        "outputId": "94e132f7-8118-4107-f2c1-9592af5bf14d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 145
        }
      },
      "source": [
        "imshow(pred_im)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f39700e9cc0>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 58
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAABvCAYAAADizAsMAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gQL9EUPDC8e5",
        "colab_type": "code",
        "outputId": "a63bd610-4364-4397-b07e-bb67c192cace",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "start_time = time.time()\n",
        "\n",
        "im = Image.open('data/demo/uu_000095.png')\n",
        "# print(im.size)\n",
        "im = im.resize((width, height), Image.ANTIALIAS)\n",
        "\n",
        "im_scaled = np.float32(im)/255.0\n",
        "\n",
        "inputs = np.array([im_scaled])\n",
        "\n",
        "label = sess.run(tf.math.softmax(output_label), feed_dict = {input_im: inputs})\n",
        "test_label = label\n",
        "#print(np.array(label).shape)\n",
        "label = label[0][:, :, 1]\n",
        "segmentation = (label > 0.5).reshape(height, width, 1)\n",
        "\n",
        "#mask = \n",
        "mask = np.dot(segmentation, np.array([[0, 255, 0, 127]]))\n",
        "mask = mask.astype('uint8')\n",
        "mask = Image.fromarray(mask, 'RGBA')\n",
        "\n",
        "im.paste(mask, mask = mask)\n",
        "\n",
        "x = np.array(im)\n",
        "\n",
        "print(\"--- %s seconds ---\" % (time.time() - start_time))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--- 1.1712534427642822 seconds ---\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BjXLJbZSy6_K",
        "colab_type": "code",
        "outputId": "f7c150b6-ec45-4a05-8aed-9a2b1e7d4006",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 145
        }
      },
      "source": [
        "imshow(x)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f97800a67f0>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 57
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAABvCAYAAADizAsMAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BC9rSm-MKJlX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oryaTKipKJiT",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "R9Tq661zKJfL",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-c-TeUHfKJXP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6UfRXTCIQJsY",
        "colab_type": "code",
        "outputId": "99763742-3af8-4fbc-9e56-8ac18142c1b7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(end_points['pool4'])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/block2/unit_4/bottleneck_v2/add:0\", shape=(?, 70, 16, 512), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b8dCBCHLkRcm",
        "colab_type": "code",
        "outputId": "6ababca8-a5b8-4604-dfc4-ec272dc500b3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "# Equivalent to pool4\n",
        "print(end_points[frontend_scope + '/block2'])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/block2/unit_4/bottleneck_v2/add:0\", shape=(?, 70, 16, 512), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4VTlJHW6fNsD",
        "colab_type": "code",
        "outputId": "fb84fbf7-6d3e-4806-cc64-ce59d649afca",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(end_points['pool5'])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/postnorm/Relu:0\", shape=(?, 35, 8, 2048), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GQuWESpmkTzN",
        "colab_type": "code",
        "outputId": "70b6ff07-82b4-4c03-ea53-e30f3081d505",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "# Equivalent to pool5\n",
        "print(logits)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/postnorm/Relu:0\", shape=(?, 35, 8, 2048), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PAYPBWGmg8DN",
        "colab_type": "code",
        "outputId": "e2a156eb-a7e3-4c6f-f794-f04877b147cb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(end_points[frontend_scope + '/block3'])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/block3/unit_23/bottleneck_v2/add:0\", shape=(?, 35, 8, 1024), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IE93RoDlkkiE",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "for key in end_points:\n",
        "  print(key)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d3kWGZrPdE-_",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "x = end_points[frontend_scope + '/block1']"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XOzH0IjXdNIb",
        "colab_type": "code",
        "outputId": "ffd21ac2-4b8a-4cf3-b38b-6c4533bbcd93",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(x)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/block1/unit_3/bottleneck_v2/add:0\", shape=(?, ?, ?, 256), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EdU7lDnhdQjs",
        "colab_type": "code",
        "outputId": "a92a24f2-1a77-40aa-8e53-2f10dbffd204",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "print(end_points['pool3'])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"resnet_v2_101/block1/unit_3/bottleneck_v2/add:0\", shape=(?, ?, ?, 256), dtype=float32)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EjCXu0A_iLyZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def main():\n",
        "  num_classes = 2\n",
        "  data_dir = './gdrive/My Drive/SemanticSegmentation/data'\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "s2b5vlrqKHfn",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "output_dir = os.path.join(save_dir, str(time.time()))\n",
        "if os.path.exists(output_dir):\n",
        "  shutil.rmtree(output_dir)\n",
        "os.makedirs(output_dir)\n",
        "\n",
        "print(\"Training finished. Saving images to {}\".format(output_dir))\n",
        "\n",
        "for image in glob(os.path.join(test_dir, 'image_2', '*.png')):\n",
        "  im = Image.open(image)\n",
        "  im = im.resize((width, height), Image.ANTIALIAS)\n",
        "\n",
        "  im_scaled = np.float32(im)/255.0\n",
        "  \n",
        "  inputs = np.array([im_scaled])\n",
        "    \n",
        "  label = sess.run(tf.math.softmax(output_label), feed_dict = {input_im: inputs})\n",
        "  test_label = label\n",
        "  #print(np.array(label).shape)\n",
        "  label = label[0][:, :, 1]\n",
        "  segmentation = (label > 0.5).reshape(height, width, 1)\n",
        "\n",
        "  #mask = \n",
        "  mask = np.dot(segmentation, np.array([[0, 255, 0, 127]]))\n",
        "  mask = mask.astype('uint8')\n",
        "  mask = Image.fromarray(mask, 'RGBA')\n",
        "\n",
        "  im.paste(mask, mask = mask)\n",
        "\n",
        "  im.save(os.path.join(output_dir, os.path.basename(image)))"
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}