Brain_tumor_segmentation_TransUNET.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Segmentation Transformers for Tumor Segmentation"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The objective is to implement an architechture that blends UNETs and transformers architechtures: TransUNET, Proposed by Jieneng Chen et al in 2021. The original papaer uses it to perform multi organ segmentation, here, we'll try to do it on Lower Grade Gliomas (Data available at https://www.kaggle.com/datasets/mateuszbuda/lgg-mri-segmentation) <br>\n",
    "\n",
    "> We argue that Transformers can serve as strong encoders for medical image segmentation tasks, with the combination of U-Net to enhance finer\n",
    "> details by recovering localized spatial information. TransUNet achieves\n",
    "> superior performances to various competing methods on different medical\n",
    "> applications including multi-organ segmentation and cardiac segmentation. Code and models are available at https://github.com/Beckschen/TransUNet\n",
    "<br>\n",
    "<br>\n",
    "<p align='center'>\n",
    "    <img src='images/sc01.jpg'/>\n",
    "</p>"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 0. libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow_io as tfio\n",
    "import numpy as np \n",
    "import pandas as pd \n",
    "import PIL\n",
    "import glob\n",
    "import os\n",
    "import matplotlib.pyplot as plt\n",
    "import random\n",
    "from tqdm import tqdm\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "%matplotlib inline\n",
    "plt.style.use('dark_background')\n",
    "\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.models import Model\n",
    "from keras import layers\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras import backend as K\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num GPUs Available:  1\n"
     ]
    }
   ],
   "source": [
    "tf.config.experimental.set_memory_growth( tf.config.list_physical_devices()[1], enable=True)\n",
    "print(\"Num GPUs Available: \", len(tf.config.list_physical_devices('GPU')))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Example-Label pairing\n",
    "\n",
    "Identify examples and labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dataset features a total of:  3929  labelled examples\n"
     ]
    }
   ],
   "source": [
    "# get the images file paths\n",
    "filenames = []\n",
    "for path, currentDirectory, files in os.walk(\"lgg-mri-segmentation\"):\n",
    "    for file in files:\n",
    "        if file.endswith('.tif'):\n",
    "            filenames.append(path + '/' + file)\n",
    "filenames = sorted(filenames)\n",
    "\n",
    "original = []\n",
    "mask = []\n",
    "for image in filenames:\n",
    "    if image.endswith('mask.tif'):\n",
    "        mask.append(image)\n",
    "    else:\n",
    "        original.append(image)\n",
    "\n",
    "mask = sorted(mask, key = len)\n",
    "original = sorted(original, key = len)\n",
    "data = pd.DataFrame({'original': original, 'mask':mask})\n",
    "data.head(10)\n",
    "\n",
    "nSamples = len(original)\n",
    "print('dataset features a total of: ',nSamples,' labelled examples')"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load and show a train example"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "each example has dimension:  (256, 256)\n",
      "each array has shape:     (256, 256, 3)\n"
     ]
    }
   ],
   "source": [
    "idx = np.random.randint(0,nSamples-1)\n",
    "img = PIL.Image.open(data['original'][idx])\n",
    "lbl = PIL.Image.open(data['mask'][idx])\n",
    "\n",
    "fig, axs = plt.subplots(1,3)\n",
    "axs[0].imshow(img), axs[0].set_title('original')\n",
    "axs[1].imshow(lbl), axs[1].set_title('label')\n",
    "axs[2].imshow(img), axs[2].set_title('manual segmentation')\n",
    "axs[2].imshow(lbl, alpha=0.5 * (np.array(lbl) > 0))\n",
    "fig.suptitle('DATA', x=0.525, y=0.8, fontsize=24)\n",
    "plt.show()\n",
    "\n",
    "img_size = np.array(img).shape[:2]\n",
    "img_shape= np.array(img).shape\n",
    "\n",
    "print('each example has dimension: ', img_size)\n",
    "print('each array has shape:    ', img_shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train, x_test = train_test_split(data, test_size = 0.2, shuffle=True)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Data loading, preprocessing and augmentation using Image Generators"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCHSIZE = 14\n",
    "def preprocess(img, mask):\n",
    "    mask[mask > 0.5] = 1\n",
    "    mask[mask <= 0.5] = 0\n",
    "    return (img, mask)\n",
    "\n",
    "def create_gen(x):\n",
    "    gen = ImageDataGenerator(\n",
    "        rescale=1./255\n",
    "    )\n",
    "\n",
    "    img_gen = gen.flow_from_dataframe(pd.DataFrame(x), \n",
    "                                      x_col='original',\n",
    "                                      class_mode=None,\n",
    "                                      horizontal_flip=True,\n",
    "                                      vertical_flip=True, \n",
    "                                      rotation_range=360,\n",
    "                                      target_size=img_size,\n",
    "                                      batch_size=BATCHSIZE,\n",
    "                                      shuffle=False\n",
    "                                     )\n",
    "    \n",
    "    mask_gen = gen.flow_from_dataframe(pd.DataFrame(x), \n",
    "                                      x_col='mask',\n",
    "                                      class_mode=None, \n",
    "                                      horizontal_flip=True,\n",
    "                                      vertical_flip=True, \n",
    "                                      rotation_range=360,\n",
    "                                      target_size=img_size,\n",
    "                                      batch_size=BATCHSIZE,\n",
    "                                      shuffle=False\n",
    "                                     )\n",
    "    generator = zip(img_gen, mask_gen)\n",
    "    \n",
    "    for (img, mask) in generator:\n",
    "        img, mask = preprocess(img, mask)\n",
    "        yield (img, mask)\n",
    "\n",
    "train = create_gen(x_train)\n",
    "val = create_gen(x_test)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Model creation: Transformer "
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.1: Patch encoding\n",
    "The first transformer block recieves the feature maps coming out from the Convolutional layers. We can interpret the activation of the filters as some sort of patches which we can enhance by applying attention. The Vision Transformers works well with (hidden) embeddings as input so what we are required to do is:\n",
    " - flatten the \"patches\";\n",
    " - apply a linear projection in a space of dimension `d_model`;\n",
    " - convert them to embeddings;\n",
    " - add the positional encoding;"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's define the `Positional Encoding` first. When the first Transformer architecture was presented with the paper 'Attention is all you need' they proposed an encoding based on sinusoidal waves. They argued that due to the linear nature of the transformations inside a transformer, an encoding of this kind would be more suited. In Vision Transformers, instead, the embeddings are initialized randomly:\n",
    "> Other than that, the position embeddings at initialization time carry no information <br>\n",
    "> about the 2D positions of the patches and all spatial relations between the patches have to be learned<br>\n",
    "> from scratch. <br>\n",
    "> [an image is worth 16x16 words: transformers for image recoginition at scale](https://arxiv.org/abs/2010.11929)\n",
    "\n",
    "Here we are still going to use randomly initialized embeddings, however we are not going to use inductive bias, since a binary segmentation problem."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The Class ``PatchEncoder`` will convert the feature maps in embeddings with positional encoding added\n",
    "\n",
    "> We map the vectorized patches xp into a latent D-dimensional embedding space using a trainable linear projection.<br>\n",
    "> [TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation](https://arxiv.org/pdf/2102.04306.pdf)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "class PatchEncoder(layers.Layer):\n",
    "    def __init__(self, num_patches, d_model):\n",
    "        super().__init__()\n",
    "        self.d_model = d_model\n",
    "        self.num_patches = num_patches\n",
    "        self.projection = layers.Dense(units=d_model)                                      \n",
    "        self.position_embedding = layers.Embedding(input_dim=num_patches, output_dim=d_model)   \n",
    "        self.flatten = layers.Reshape((num_patches,-1)) \n",
    "\n",
    "    def call(self, x): \n",
    "        x = self.flatten(x)                                                                                    \n",
    "        positions = tf.range(start=0, limit=self.num_patches, delta=1)\n",
    "        x = self.projection(x)                                          \n",
    "        x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))                          \n",
    "        encoded = x + self.position_embedding(positions)\n",
    "        return encoded"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "for comparison, in NLP Transformers, the embedding shape was: `(batch_size) x (query_length) x (d_model)` <br>\n",
    "here, we are dealing with a query length equals to n_pathces so: `(batch_size) x (n_patches) x (d_model)` <br>\n",
    "Dimensions check out.<br>\n",
    "[Neural machine translation with a Transformer and Keras](https://www.tensorflow.org/text/tutorials/transformer)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.2 attention"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "class BaseAttention(tf.keras.layers.Layer):\n",
    "  def __init__(self, **kwargs):\n",
    "    super().__init__()\n",
    "    self.mha = tf.keras.layers.MultiHeadAttention(**kwargs)\n",
    "    self.layernorm = tf.keras.layers.LayerNormalization()\n",
    "    self.add = tf.keras.layers.Add()\n",
    "\n",
    "class GlobalSelfAttention(BaseAttention):\n",
    "  def __init__(self, **kwargs):\n",
    "    super().__init__(**kwargs)\n",
    "    self.layer_output = None\n",
    "\n",
    "  def call(self, x):\n",
    "    attn_output, attn_scores = self.mha(\n",
    "        query=x,\n",
    "        value=x,\n",
    "        key=x,\n",
    "        return_attention_scores=True)\n",
    "     \n",
    "    x = self.add([x, attn_output])\n",
    "    x = self.layernorm(x)\n",
    "    return x"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.3 Multi Layer Perceptron layer\n",
    "We'll adhere to the original study:\n",
    "> For the “base” model, the hidden\n",
    "> size D, number of layers, MLP size, and number of heads are set to be 12, 768,\n",
    "> 3072, and 12, respectively while those hyperparamters for “large” model are 24,\n",
    "> 1024, 4096, and 16.<br>\n",
    "> [TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation](https://arxiv.org/pdf/2102.04306.pdf)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "class FeedForward(tf.keras.layers.Layer):\n",
    "  def __init__(self, d_model, dff, dropout_rate=0.1):\n",
    "    super().__init__()\n",
    "    self.layer_output = None\n",
    "    self.seq = tf.keras.Sequential([\n",
    "      tf.keras.layers.Dense(dff, activation='relu'),\n",
    "      tf.keras.layers.Dense(d_model),\n",
    "      tf.keras.layers.Dropout(dropout_rate)\n",
    "    ])\n",
    "    self.add = tf.keras.layers.Add()\n",
    "    self.layer_norm = tf.keras.layers.LayerNormalization()\n",
    "    \n",
    "  def call(self, x):\n",
    "    x = self.add([x, self.seq(x)])\n",
    "    x = self.layer_norm(x) \n",
    "    self.layer_output = x\n",
    "\n",
    "    return x"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.4 Transformer block\n",
    "The authors work with pretrained ViT. We'll try to train them from scratch instead. \n",
    "> For the hybrid encoder design, we combine ResNet-50 and ViT,\n",
    "> denoted as “R50-ViT”, throught this paper. All Transformer backbones (i.e.,\n",
    "> ViT) and ResNet-50 (denoted as “R-50”) were pretrained on ImageNet <br>\n",
    "[TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation](https://arxiv.org/pdf/2102.04306.pdf)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "class EncoderLayer(tf.keras.layers.Layer):\n",
    "  def __init__(self,*, d_model, num_heads, dff, dropout_rate=0.1):\n",
    "    super().__init__()\n",
    "\n",
    "    self.self_attention = GlobalSelfAttention(\n",
    "        num_heads=num_heads,\n",
    "        key_dim=d_model,\n",
    "        dropout=dropout_rate)\n",
    "\n",
    "    self.ffn = FeedForward(d_model, dff)\n",
    "  \n",
    "  def call(self, x):\n",
    "    x = self.self_attention(x)\n",
    "    x = self.ffn(x)\n",
    "    \n",
    "    return x\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model Hyper-Parameters\n",
    "\n",
    "Using parameters similar to the original study:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_tf_layers = 4 # original got 12 but they don't fit\n",
    "dff = 3072\n",
    "d_model = 768 \n",
    "num_heads = 8\n",
    "n_classes = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_input = layers.Input(img_shape)\n",
    "\n",
    "# Block 1\n",
    "x = layers.Conv2D(64, (3, 3), padding='same', name='block1_conv1')(img_input)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "x = layers.MaxPooling2D()(x)\n",
    "block_1_out = x\n",
    "\n",
    "# Block 2\n",
    "x = layers.Conv2D(128, (3, 3), padding='same', name='block2_conv1')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "x = layers.MaxPooling2D()(x)\n",
    "block_2_out = x\n",
    "\n",
    "# Block 3\n",
    "x = layers.Conv2D(256, (3, 3), padding='same', name='block3_conv1')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "x = layers.MaxPooling2D()(x)\n",
    "block_3_out = x\n",
    "\n",
    "# Transformer Blocks\n",
    "x = PatchEncoder(num_patches=x.shape[-1], d_model=d_model)(x)\n",
    "for _ in range(n_tf_layers):\n",
    "    x = EncoderLayer(d_model=d_model, dff=dff, num_heads=num_heads)(x)\n",
    "\n",
    "# Channel Size Reduction\n",
    "x = layers.Reshape((16, 16, -1))(x)\n",
    "x = layers.Conv2D(512, (1, 1), padding='same', name='block_hid_conv1')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "\n",
    "# Up Sampling 1\n",
    "x = layers.Conv2DTranspose(256, (2, 2), strides=(2, 2), padding='same', name = 'Conv2DTranspose_UP1')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "x = layers.concatenate([x, block_3_out])\n",
    "x = layers.Conv2D(256, (3, 3), padding='same')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "\n",
    "# Up Sampling 2\n",
    "x = layers.Conv2DTranspose(128, (2, 2), strides=(2, 2), padding='same', name = 'Conv2DTranspose_UP2')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "x = layers.concatenate([x, block_2_out])\n",
    "x = layers.Conv2D(128, (3, 3), padding='same')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "\n",
    "\n",
    "# Up Sampling 3\n",
    "x = layers.Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same', name = 'Conv2DTranspose_UP3')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "x = layers.concatenate([x, block_1_out])\n",
    "x = layers.Conv2D(64, (3, 3), padding='same')(x)\n",
    "x = layers.BatchNormalization()(x)\n",
    "x = layers.Activation('relu')(x)\n",
    "\n",
    "# Up Sampling 4\n",
    "x = layers.Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same', name = 'Conv2DTranspose_UP4')(x)\n",
    "x = layers.Conv2D(n_classes, (3, 3), activation='sigmoid', padding='same')(x)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Model Training\n",
    "\n",
    "There are some options here. For transformers architecture, a dynamic learning rate along with early stopping were first chosen:\n",
    "> We train all models, including ResNets, using Adam (Kingma & Ba, 2015) with β1 = 0.9, β2 = 0.999, a batch size of 4096 and apply a high weight decay of 0.1, whic we found to be useful for transfer of all models (Appendix D.1 shows that, in contrast to common practices, Adam works slightly better than SGD for ResNets in our setting). We use a linear learning rate warmup and decay.<br>\n",
    "> [an image is worth 16x16 words: transformers for image recoginition at scale](https://arxiv.org/abs/2010.11929)\n",
    "\n",
    "\n",
    "\n",
    "However, the authors of the TransUnet use a static strategy:\n",
    "> Models are trained with SGD optimizer with learning rate 0.01, momentum 0.9 and weight decay 1e-4. The\n",
    "> default batch size is 24 and the default number of training iterations are 20k<br>\n",
    "> [TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation](https://arxiv.org/pdf/2102.04306.pdf)\n",
    "\n",
    "We'll try to stick to the TransUNET study"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):\n",
    "  def __init__(self, d_model, warmup_steps=4000):\n",
    "    super().__init__()\n",
    "\n",
    "    self.d_model = d_model\n",
    "    self.d_model = tf.cast(self.d_model, tf.float32)\n",
    "\n",
    "    self.warmup_steps = warmup_steps\n",
    "\n",
    "  def __call__(self, step):\n",
    "    step = tf.cast(step, dtype=tf.float32)\n",
    "    arg1 = tf.math.rsqrt(step)\n",
    "    arg2 = step * (self.warmup_steps ** -1.5)\n",
    "\n",
    "    return tf.math.rsqrt(self.d_model) * tf.math.minimum(arg1, arg2)\n",
    "\n",
    "learning_rate = CustomSchedule(d_model)\n",
    "\n",
    "optimizer_ViT = tf.keras.optimizers.Adam(learning_rate, beta_1=0.9, beta_2=0.98,\n",
    "                                     epsilon=1e-9)\n",
    "\n",
    "plt.plot(learning_rate(tf.range(40000, dtype=tf.float32)))\n",
    "plt.ylabel('Learning Rate')\n",
    "plt.xlabel('Train Step')\n",
    "plt.show()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here is the adaptive learning rate mentioned before if you want to experiment with it."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizer_transunet = tf.keras.optimizers.SGD(learning_rate=0.01, \n",
    "                                              momentum=0.9)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_1\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_2 (InputLayer)           [(None, 256, 256, 3  0           []                               \n",
      "                                )]                                                                \n",
      "                                                                                                  \n",
      " block1_conv1 (Conv2D)          (None, 256, 256, 64  1792        ['input_2[0][0]']                \n",
      "                                )                                                                 \n",
      "                                                                                                  \n",
      " batch_normalization_10 (BatchN  (None, 256, 256, 64  256        ['block1_conv1[0][0]']           \n",
      " ormalization)                  )                                                                 \n",
      "                                                                                                  \n",
      " activation_9 (Activation)      (None, 256, 256, 64  0           ['batch_normalization_10[0][0]'] \n",
      "                                )                                                                 \n",
      "                                                                                                  \n",
      " max_pooling2d_3 (MaxPooling2D)  (None, 128, 128, 64  0          ['activation_9[0][0]']           \n",
      "                                )                                                                 \n",
      "                                                                                                  \n",
      " block2_conv1 (Conv2D)          (None, 128, 128, 12  73856       ['max_pooling2d_3[0][0]']        \n",
      "                                8)                                                                \n",
      "                                                                                                  \n",
      " batch_normalization_11 (BatchN  (None, 128, 128, 12  512        ['block2_conv1[0][0]']           \n",
      " ormalization)                  8)                                                                \n",
      "                                                                                                  \n",
      " activation_10 (Activation)     (None, 128, 128, 12  0           ['batch_normalization_11[0][0]'] \n",
      "                                8)                                                                \n",
      "                                                                                                  \n",
      " max_pooling2d_4 (MaxPooling2D)  (None, 64, 64, 128)  0          ['activation_10[0][0]']          \n",
      "                                                                                                  \n",
      " block3_conv1 (Conv2D)          (None, 64, 64, 256)  295168      ['max_pooling2d_4[0][0]']        \n",
      "                                                                                                  \n",
      " batch_normalization_12 (BatchN  (None, 64, 64, 256)  1024       ['block3_conv1[0][0]']           \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_11 (Activation)     (None, 64, 64, 256)  0           ['batch_normalization_12[0][0]'] \n",
      "                                                                                                  \n",
      " max_pooling2d_5 (MaxPooling2D)  (None, 32, 32, 256)  0          ['activation_11[0][0]']          \n",
      "                                                                                                  \n",
      " patch_encoder_1 (PatchEncoder)  (None, 256, 768)    983808      ['max_pooling2d_5[0][0]']        \n",
      "                                                                                                  \n",
      " encoder_layer_4 (EncoderLayer)  (None, 256, 768)    23619072    ['patch_encoder_1[0][0]']        \n",
      "                                                                                                  \n",
      " encoder_layer_5 (EncoderLayer)  (None, 256, 768)    23619072    ['encoder_layer_4[0][0]']        \n",
      "                                                                                                  \n",
      " encoder_layer_6 (EncoderLayer)  (None, 256, 768)    23619072    ['encoder_layer_5[0][0]']        \n",
      "                                                                                                  \n",
      " encoder_layer_7 (EncoderLayer)  (None, 256, 768)    23619072    ['encoder_layer_6[0][0]']        \n",
      "                                                                                                  \n",
      " reshape_3 (Reshape)            (None, 16, 16, 768)  0           ['encoder_layer_7[0][0]']        \n",
      "                                                                                                  \n",
      " block_hid_conv1 (Conv2D)       (None, 16, 16, 512)  393728      ['reshape_3[0][0]']              \n",
      "                                                                                                  \n",
      " batch_normalization_13 (BatchN  (None, 16, 16, 512)  2048       ['block_hid_conv1[0][0]']        \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " Conv2DTranspose_UP1 (Conv2DTra  (None, 32, 32, 256)  524544     ['batch_normalization_13[0][0]'] \n",
      " nspose)                                                                                          \n",
      "                                                                                                  \n",
      " batch_normalization_14 (BatchN  (None, 32, 32, 256)  1024       ['Conv2DTranspose_UP1[0][0]']    \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_12 (Activation)     (None, 32, 32, 256)  0           ['batch_normalization_14[0][0]'] \n",
      "                                                                                                  \n",
      " concatenate_3 (Concatenate)    (None, 32, 32, 512)  0           ['activation_12[0][0]',          \n",
      "                                                                  'max_pooling2d_5[0][0]']        \n",
      "                                                                                                  \n",
      " conv2d_4 (Conv2D)              (None, 32, 32, 256)  1179904     ['concatenate_3[0][0]']          \n",
      "                                                                                                  \n",
      " batch_normalization_15 (BatchN  (None, 32, 32, 256)  1024       ['conv2d_4[0][0]']               \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_13 (Activation)     (None, 32, 32, 256)  0           ['batch_normalization_15[0][0]'] \n",
      "                                                                                                  \n",
      " Conv2DTranspose_UP2 (Conv2DTra  (None, 64, 64, 128)  131200     ['activation_13[0][0]']          \n",
      " nspose)                                                                                          \n",
      "                                                                                                  \n",
      " batch_normalization_16 (BatchN  (None, 64, 64, 128)  512        ['Conv2DTranspose_UP2[0][0]']    \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_14 (Activation)     (None, 64, 64, 128)  0           ['batch_normalization_16[0][0]'] \n",
      "                                                                                                  \n",
      " concatenate_4 (Concatenate)    (None, 64, 64, 256)  0           ['activation_14[0][0]',          \n",
      "                                                                  'max_pooling2d_4[0][0]']        \n",
      "                                                                                                  \n",
      " conv2d_5 (Conv2D)              (None, 64, 64, 128)  295040      ['concatenate_4[0][0]']          \n",
      "                                                                                                  \n",
      " batch_normalization_17 (BatchN  (None, 64, 64, 128)  512        ['conv2d_5[0][0]']               \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_15 (Activation)     (None, 64, 64, 128)  0           ['batch_normalization_17[0][0]'] \n",
      "                                                                                                  \n",
      " Conv2DTranspose_UP3 (Conv2DTra  (None, 128, 128, 64  32832      ['activation_15[0][0]']          \n",
      " nspose)                        )                                                                 \n",
      "                                                                                                  \n",
      " batch_normalization_18 (BatchN  (None, 128, 128, 64  256        ['Conv2DTranspose_UP3[0][0]']    \n",
      " ormalization)                  )                                                                 \n",
      "                                                                                                  \n",
      " activation_16 (Activation)     (None, 128, 128, 64  0           ['batch_normalization_18[0][0]'] \n",
      "                                )                                                                 \n",
      "                                                                                                  \n",
      " concatenate_5 (Concatenate)    (None, 128, 128, 12  0           ['activation_16[0][0]',          \n",
      "                                8)                                'max_pooling2d_3[0][0]']        \n",
      "                                                                                                  \n",
      " conv2d_6 (Conv2D)              (None, 128, 128, 64  73792       ['concatenate_5[0][0]']          \n",
      "                                )                                                                 \n",
      "                                                                                                  \n",
      " batch_normalization_19 (BatchN  (None, 128, 128, 64  256        ['conv2d_6[0][0]']               \n",
      " ormalization)                  )                                                                 \n",
      "                                                                                                  \n",
      " activation_17 (Activation)     (None, 128, 128, 64  0           ['batch_normalization_19[0][0]'] \n",
      "                                )                                                                 \n",
      "                                                                                                  \n",
      " Conv2DTranspose_UP4 (Conv2DTra  (None, 256, 256, 32  8224       ['activation_17[0][0]']          \n",
      " nspose)                        )                                                                 \n",
      "                                                                                                  \n",
      " conv2d_7 (Conv2D)              (None, 256, 256, 1)  289         ['Conv2DTranspose_UP4[0][0]']    \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 98,477,889\n",
      "Trainable params: 98,474,177\n",
      "Non-trainable params: 3,712\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Model(img_input, x)\n",
    "model.compile(optimizer=optimizer_transunet, loss='binary_crossentropy')\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Fit\n",
    "Due to the number of parameters and the fact we are training everything from scratch, we are going to increase the training epochs. We'll se where this brings us.<br>\n",
    "*Note: when increasing the number of epochs, one should implement also call backs measure for preacustion*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 3143 validated image filenames.\n",
      "Found 3143 validated image filenames.\n",
      "Epoch 1/2000\n",
      "224/224 [==============================] - ETA: 0s - loss: 0.0521Found 786 validated image filenames.\n",
      "Found 786 validated image filenames.\n",
      "224/224 [==============================] - 61s 259ms/step - loss: 0.0521 - val_loss: 0.0504\n",
      "Epoch 2/2000\n",
      "224/224 [==============================] - 59s 259ms/step - loss: 0.0251 - val_loss: 0.0390\n",
      "Epoch 3/2000\n",
      "224/224 [==============================] - 58s 259ms/step - loss: 0.0199 - val_loss: 0.0189\n",
      "Epoch 4/2000\n",
      "224/224 [==============================] - 57s 257ms/step - loss: 0.0166 - val_loss: 0.0158\n",
      "Epoch 5/2000\n",
      "224/224 [==============================] - 58s 258ms/step - loss: 0.0143 - val_loss: 0.0131\n",
      "Epoch 6/2000\n",
      "224/224 [==============================] - 58s 258ms/step - loss: 0.0124 - val_loss: 0.0128\n",
      "Epoch 7/2000\n",
      "224/224 [==============================] - 57s 254ms/step - loss: 0.0108 - val_loss: 0.0109\n",
      "Epoch 8/2000\n",
      "224/224 [==============================] - 56s 252ms/step - loss: 0.0095 - val_loss: 0.0104\n",
      "Epoch 9/2000\n",
      "224/224 [==============================] - 57s 254ms/step - loss: 0.0086 - val_loss: 0.0100\n",
      "Epoch 10/2000\n",
      "224/224 [==============================] - 57s 255ms/step - loss: 0.0081 - val_loss: 0.0100\n",
      "Epoch 11/2000\n",
      "224/224 [==============================] - 57s 255ms/step - loss: 0.0075 - val_loss: 0.0087\n",
      "Epoch 12/2000\n",
      "224/224 [==============================] - 57s 254ms/step - loss: 0.0069 - val_loss: 0.0091\n",
      "Epoch 13/2000\n",
      "224/224 [==============================] - 58s 257ms/step - loss: 0.0065 - val_loss: 0.0088\n",
      "Epoch 14/2000\n",
      "224/224 [==============================] - 57s 254ms/step - loss: 0.0062 - val_loss: 0.0090\n",
      "Epoch 15/2000\n",
      "224/224 [==============================] - 57s 257ms/step - loss: 0.0060 - val_loss: 0.0091\n",
      "Epoch 16/2000\n",
      "224/224 [==============================] - 57s 256ms/step - loss: 0.0058 - val_loss: 0.0084\n",
      "Epoch 17/2000\n",
      "224/224 [==============================] - 57s 254ms/step - loss: 0.0055 - val_loss: 0.0087\n",
      "Epoch 18/2000\n",
      "224/224 [==============================] - 58s 257ms/step - loss: 0.0053 - val_loss: 0.0102\n",
      "Epoch 19/2000\n",
      "224/224 [==============================] - 57s 256ms/step - loss: 0.0052 - val_loss: 0.0101\n",
      "Epoch 20/2000\n",
      "224/224 [==============================] - 57s 255ms/step - loss: 0.0050 - val_loss: 0.0084\n",
      "Epoch 21/2000\n",
      "224/224 [==============================] - 57s 255ms/step - loss: 0.0049 - val_loss: 0.0095\n",
      "Epoch 22/2000\n",
      "224/224 [==============================] - 58s 257ms/step - loss: 0.0047 - val_loss: 0.0102\n",
      "Epoch 23/2000\n",
      "224/224 [==============================] - 58s 257ms/step - loss: 0.0047 - val_loss: 0.0084\n",
      "Epoch 24/2000\n",
      "224/224 [==============================] - ETA: 0s - loss: 0.0045"
     ]
    }
   ],
   "source": [
    "history = model.fit(train, \n",
    "          epochs=2000, \n",
    "          validation_data= val,\n",
    "          steps_per_epoch= len(x_train)// BATCHSIZE,\n",
    "          validation_steps= len(x_test) // BATCHSIZE,\n",
    "          verbose=1\n",
    "          )"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 5. Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.plot(history.history['loss'], label = 'loss')\n",
    "plt.plot(history.history['val_loss'], label = 'val_loss')\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('loss')\n",
    "plt.legend(loc='upper right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "if not(os.path.exists('TransUNET_output')):\n",
    "    os.mkdir('TransUNET_output')\n",
    "\n",
    "nSaveImages = 50\n",
    "for im in range(nSaveImages):\n",
    "    rows = 3\n",
    "    cols = 4\n",
    "    fig, axs = plt.subplots(rows, cols, figsize = (60, 60))    \n",
    "\n",
    "    for i in range(rows):\n",
    "\n",
    "        idx = np.random.randint(0,nSamples-1)\n",
    "        img = PIL.Image.open(data['original'][idx])\n",
    "        lbl = PIL.Image.open(data['mask'][idx])\n",
    "        \n",
    "        \n",
    "        pred = model.predict(np.expand_dims(np.array(img)/255,0))\n",
    "        \n",
    "        axs[i,0].imshow(img), axs[i,0].set_title('original',fontsize=50)\n",
    "        axs[i,1].imshow(lbl), axs[i,1].set_title('label',fontsize=50)\n",
    "\n",
    "        axs[i,2].imshow(img), axs[i,2].set_title('automatic segmentation', fontsize=50)\n",
    "        axs[i,2].imshow(pred[0], alpha=((pred[0] > 0.1) * 0.5)[:,:,0])\n",
    "\n",
    "        axs[i,3].imshow(pred[0], cmap='gray'), axs[i,3].set_title('predicted',fontsize=50)\n",
    "        \n",
    "    fig.suptitle('MODEL EVALUATION', fontsize=50)\n",
    "    import uuid\n",
    "    figname = uuid.uuid1()\n",
    "    fig.savefig('TransUNET_output/'+str(figname))\n",
    "    plt.close()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "autoencoders",
   "language": "python",
   "name": "autoencoders"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.0"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}