DeepCNNmodel_Modified.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "6c1666e0",
   "metadata": {},
   "outputs": [],
   "source": [
    "def DeepCNNmodel_Modified():\n",
    "# def DeepCNNmodel_Modified(LastLayer_info):\n",
    "\n",
    "    \n",
    "    imageSize = (65, 65, 3)\n",
    "    num_classes = 3\n",
    "    \n",
    "    input_layer = Input(shape=imageSize, name='I1')\n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv1_1')(input_layer)\n",
    "    x = BatchNormalization(name='B1_1')(x)\n",
    "    relu1_1 = Activation('relu', name='relu1_1')(x)\n",
    "    \n",
    "    \n",
    "    \n",
    "    x1 = Conv2D(64, (5, 5), padding='same', name='Conv1_2')(relu1_1)\n",
    "    x1 = BatchNormalization(name='B1_2')(x1)\n",
    "    x1 = Activation('relu', name='relu1_2')(x1)\n",
    "    \n",
    "    x2 = Conv2D(64, (7, 7), padding='same', name='Conv1_3')(x1)\n",
    "    x2 = BatchNormalization(name='B1_3')(x2) \n",
    "    relu1_3 = Activation('relu', name='relu1_3')(x2)\n",
    "     \n",
    "    \n",
    "    x_depthcat_2 = Concatenate(name='depthcat_2')([relu1_1, relu1_3])            #depthcat_2\n",
    "    x = Conv2D(64, (9, 9), padding='same', name='Conv1_4')(x_depthcat_2)\n",
    "    x = BatchNormalization(name='B1_4')(x)\n",
    "    x = Activation('relu', name='relu1_4')(x)\n",
    "    \n",
    "    x1 = Conv2D(64, (11, 11), padding='same',name='Conv1_5')(x)\n",
    "    x1 = BatchNormalization(name='B1_5')(x1)\n",
    "    relu1_5 = Activation('relu', name='relu1_5')(x1)\n",
    "    \n",
    "    depthcat_3 = Concatenate(name='depthcat_3')([relu1_5, x_depthcat_2])         #depthcat_3\n",
    "    x = Conv2D(64, (13, 13), padding='same', name='Conv1_6')(depthcat_3)\n",
    "    x = BatchNormalization(name='B1_6')(x)\n",
    "    relu1_6 = Activation('relu', name='relu1_6')(x)\n",
    "    \n",
    "    x1 = Conv2D(64, (15, 15), padding='same', name='Conv1_7')(relu1_6)\n",
    "    x1 = BatchNormalization(name='B1_7')(x1)\n",
    "    relu1_7 = Activation('relu', name='relu1_7')(x1)\n",
    "    \n",
    "    depthcat_4 = Concatenate(name='depthcat_4')([relu1_7, depthcat_3])            #depthcat_4\n",
    "    x = Conv2D(64, (17, 17), padding='same', name='Conv1_8')(depthcat_4)\n",
    "    x = BatchNormalization(name='B1_8')(x)\n",
    "    x = Activation('relu', name='relu1_8')(x)\n",
    "    \n",
    "    x1 = Conv2D(64, (19, 19), padding='same', name='Conv1_9')(x)\n",
    "    x = BatchNormalization(name='B1_9')(x1)\n",
    "    relu1_9 = Activation('relu', name='relu1_9')(x1)\n",
    "    \n",
    "    depthcat_1 = Concatenate(name='depthcat_1')([relu1_9, depthcat_4])            #depthcat_1\n",
    "    x = Conv2D(64, (21, 21), padding='same', name='Conv1_10')(depthcat_1)\n",
    "    #print(x.shape)\n",
    "    x = BatchNormalization(name='B1_10')(x)    \n",
    "    x = Activation('relu', name='relu1_10')(x)\n",
    "    #print(x.shape)\n",
    "    D1_1 = Dropout(0.5, name='D1_1')(x) \n",
    "    #print(D1_1.shape)\n",
    "    #D1_1 = Conv2D(64, 2, strides=1,padding='same', name='D1_1_deshaped')(D1_1) \n",
    "    D1_1 = Conv2D(64, 2, strides=1,padding='valid', name='D1_1_deshaped')(D1_1) \n",
    "    #print(D1_1.shape)\n",
    "    \n",
    "    \n",
    "    \n",
    "    \n",
    "    ######################################\n",
    "    ######################################\n",
    "    ######################################\n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_1')(relu1_1)\n",
    "    x = BatchNormalization(name='B2_1')(x)\n",
    "    x = Activation('relu', name='relu2_1')(x)\n",
    "    Max2_1 = MaxPooling2D((2,2), strides=(2, 2), name='Max2_1')(x)\n",
    "    #print(Max2_1.shape)\n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_2')(Max2_1)\n",
    "    #print(x.shape)\n",
    "    x = BatchNormalization(name='B2_2')(x)\n",
    "    x = Activation('relu', name='relu2_2')(x)\n",
    "    Max2_2 = MaxPooling2D((2,2), strides=(2, 2), name='Max2_2')(x)\n",
    "    \n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_3')(Max2_2)\n",
    "    x = BatchNormalization(name='B2_3')(x)\n",
    "    x = Activation('relu', name='relu2_3')(x)\n",
    "    Max2_3 = MaxPooling2D((2, 2), strides=(2, 2), name='Max2_3')(x)\n",
    "    \n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_4')(Max2_3)\n",
    "    x = BatchNormalization(name='B2_4')(x)\n",
    "    x = Activation('relu', name='relu2_4')(x)\n",
    "    Max2_4 = MaxPooling2D((2, 2), strides=(2, 2), name='Max2_4')(x)\n",
    "    x = UpSampling2D(name='unpool2_4')(Max2_4)       \n",
    "    #x = MaxUnpooling2D(name='unpool2_4')([x, MaxPooling2D((2, 2), strides=(2, 2))(x)])\n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_5')(x)\n",
    "    x = BatchNormalization(name='B2_5')(x)\n",
    "    relu2_5 = Activation('relu', name='relu2_5')(x)\n",
    "    #############################################################\n",
    "    #############################################################\n",
    "    \n",
    "    x = concatenate([relu2_5, Max2_3], axis=-1)    # UpSampling2D layer expects only one input tensor\n",
    "    x = UpSampling2D(name='unpool2_3')(x)\n",
    "    #x = MaxUnpooling2D(name='unpool2_3')([x, MaxPooling2D((2, 2), strides=(2, 2))(x)])\n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_6')(x)\n",
    "    x = BatchNormalization(name='B2_6')(x)\n",
    "    relu2_6 = Activation('relu', name='relu2_6')(x)\n",
    "    \n",
    "    x = concatenate([relu2_6, Max2_2], axis=-1)    # UpSampling2D layer expects only one input tensor\n",
    "    x = UpSampling2D(name='unpool2_2')(x)\n",
    "    #x = MaxUnpooling2D(name='unpool2_2')([x, MaxPooling2D((2, 2), strides=(2, 2))(x)])\n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_7')(x)\n",
    "    x = BatchNormalization(name='B2_7')(x)\n",
    "    relu2_7 = Activation('relu', name='relu2_7')(x)\n",
    "    \n",
    "    x = concatenate([relu2_7, Max2_1], axis=-1)    # UpSampling2D layer expects only one input tensor    \n",
    "    x = UpSampling2D(name='unpool2_1')(x)\n",
    "    #x = MaxUnpooling2D(name='unpool2_1')([x, MaxPooling2D((2, 2), strides=(2, 2))(x)])\n",
    "    \n",
    "    x = Conv2D(64, (3, 3), padding='same', name='Conv2_8')(x)\n",
    "    x = BatchNormalization(name='B2_8')(x)\n",
    "    x = Activation('relu', name='relu2_8')(x)\n",
    "    D2_1 = Dropout(0.5, name='D2_1')(x)\n",
    "    \n",
    "    ################################\n",
    "    ################################\n",
    "    ################################\n",
    "    \n",
    "    \n",
    "    relu1_1_deshaped = Conv2D(64, 2, strides=1, name='relu1_1_conv')(relu1_1)    \n",
    "    \n",
    "    print(relu1_1_deshaped.shape, 'relu1_1 shape after passing conve 2*2')\n",
    "    print(D1_1.shape, 'D1_1 shape after passing conve 2*2')\n",
    "    print(D2_1.shape)\n",
    "    \n",
    "    \n",
    "    \n",
    "    #x = Concatenate(name='depthcat_5')([D1_1, D2_1, relu1_1_deshaped])              #depthcat_5\n",
    "    #print(x.shape)\n",
    "    \n",
    "    \n",
    "    #Add padding to the input tensors\n",
    "    D1_1_padded = ZeroPadding2D(padding=((1, 0), (1, 0)), name='D1_1_padded')(D1_1)\n",
    "    D2_1_padded = ZeroPadding2D(padding=((1, 0), (1, 0)), name='D2_1_padded')(D2_1)\n",
    "    relu1_1_deshaped_padded = ZeroPadding2D(padding=((1, 0), (1, 0)), name='relu1_1_deshaped_padded')(relu1_1_deshaped)\n",
    "    \n",
    "    # Concatenate the padded tensors along the channel axis\n",
    "    x = Concatenate(name='depthcat_5')([D1_1_padded, D2_1_padded, relu1_1_deshaped_padded])\n",
    "    print(x.shape) # (None, 65, 65, 192)\n",
    "    \n",
    "    \n",
    "    \n",
    "    #x = Concatenate(name='depthcat_5')([D1_1, D2_1, relu1_1])\n",
    "    x = Conv2D(num_classes, (1, 1), padding='same', name='Conv2_class')(x)\n",
    "    output_layer = Softmax(name='Softmax')(x)\n",
    "    #output_layer = Conv2D(num_classes, (1, 1), activation='softmax', name='pixel')\n",
    "\n",
    "\n",
    "    model = Model(inputs=[input_layer], outputs=[output_layer])\n",
    "    model.summary()\n",
    "    \n",
    "    return model"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}