01-simple-regression-train.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.2.1\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU'),\n",
       " PhysicalDevice(name='/physical_device:XLA_CPU:0', device_type='XLA_CPU'),\n",
       " PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU'),\n",
       " PhysicalDevice(name='/physical_device:XLA_GPU:0', device_type='XLA_GPU')]"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "print(tf.__version__)\n",
    "tf.config.list_physical_devices()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.applications import VGG16\n",
    "from tensorflow.keras.layers import Flatten\n",
    "from tensorflow.keras.layers import Dense\n",
    "from tensorflow.keras.layers import Input\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras.preprocessing.image import img_to_array\n",
    "from tensorflow.keras.preprocessing.image import load_img\n",
    "from sklearn.model_selection import train_test_split\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import cv2\n",
    "import os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "BASE_PATH = \"dataset\"\n",
    "IMAGES_PATH = os.path.sep.join([BASE_PATH, \"images\"])\n",
    "ANNOTS_PATH = os.path.sep.join([BASE_PATH, \"airplanes.csv\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "rows = open(ANNOTS_PATH).read().strip().split(\"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# initialize the list of data (images), our target output predictions\n",
    "# (bounding box coordinates), along with the filenames of the\n",
    "# individual images\n",
    "data = []\n",
    "targets = []\n",
    "filenames = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# loop over the rows\n",
    "for row in rows:\n",
    "    # break the row into the filename and bounding box coordinates\n",
    "    (filename, startX, startY, endX, endY) = row.split(\",\")\n",
    "\n",
    "    # derive the path to the input image, load the image (in OpenCV\n",
    "    # format), and grab its dimensions\n",
    "    imagePath = os.path.sep.join([IMAGES_PATH, filename])\n",
    "    image = cv2.imread(imagePath)\n",
    "    (h, w) = image.shape[:2]\n",
    "\n",
    "    # scale the bounding box coordinates relative to the spatial\n",
    "    # dimensions of the input image\n",
    "    startX = float(startX) / w\n",
    "    startY = float(startY) / h\n",
    "    endX = float(endX) / w\n",
    "    endY = float(endY) / h\n",
    "\n",
    "    # load the image and preprocess it\n",
    "    image = load_img(imagePath, target_size=(224, 224))\n",
    "    image = img_to_array(image)\n",
    "\n",
    "    # update our list of data, targets, and filenames\n",
    "    data.append(image)\n",
    "    targets.append((startX, startY, endX, endY))\n",
    "    filenames.append(filename)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# convert the data and targets to NumPy arrays, scaling the input\n",
    "# pixel intensities from the range [0, 255] to [0, 1]\n",
    "data = np.array(data, dtype=\"float32\") / 255.0\n",
    "targets = np.array(targets, dtype=\"float32\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# partition the data into training and testing splits using 90% of\n",
    "# the data for training and the remaining 10% for testing\n",
    "split = train_test_split(data, targets, filenames, test_size=0.10, random_state=42)\n",
    "\n",
    "# unpack the data split\n",
    "(trainImages, testImages) = split[:2]\n",
    "(trainTargets, testTargets) = split[2:4]\n",
    "(trainFilenames, testFilenames) = split[4:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load the VGG16 network, ensuring the head FC layers are left off\n",
    "vgg = VGG16(weights=\"imagenet\", include_top=False, input_tensor=Input(shape=(224, 224, 3)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(vgg)\n",
    "model.add(Flatten())\n",
    "model.add( Dense(128, activation=\"relu\"))\n",
    "model.add( Dense(64, activation=\"relu\"))\n",
    "model.add( Dense(32, activation=\"relu\"))\n",
    "model.add( Dense(4, activation=\"sigmoid\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "vgg.trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "vgg16 (Model)                (None, 7, 7, 512)         14714688  \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 25088)             0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 128)               3211392   \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 4)                 132       \n",
      "=================================================================\n",
      "Total params: 17,936,548\n",
      "Trainable params: 3,221,860\n",
      "Non-trainable params: 14,714,688\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "opt = Adam(lr=1e-4)\n",
    "model.compile(loss=\"mse\", optimizer=opt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/25\n",
      "23/23 [==============================] - 24s 1s/step - loss: 0.0205 - val_loss: 0.0051\n",
      "Epoch 2/25\n",
      "23/23 [==============================] - 21s 931ms/step - loss: 0.0023 - val_loss: 8.6541e-04\n",
      "Epoch 3/25\n",
      "23/23 [==============================] - 21s 905ms/step - loss: 0.0011 - val_loss: 6.2884e-04\n",
      "Epoch 4/25\n",
      "23/23 [==============================] - 21s 928ms/step - loss: 9.4032e-04 - val_loss: 5.8869e-04\n",
      "Epoch 5/25\n",
      "23/23 [==============================] - 22s 941ms/step - loss: 7.9146e-04 - val_loss: 5.5063e-04\n",
      "Epoch 6/25\n",
      "23/23 [==============================] - 22s 941ms/step - loss: 6.9295e-04 - val_loss: 5.3969e-04\n",
      "Epoch 7/25\n",
      "23/23 [==============================] - 22s 949ms/step - loss: 5.9885e-04 - val_loss: 5.6391e-04\n",
      "Epoch 8/25\n",
      "23/23 [==============================] - 22s 957ms/step - loss: 5.1892e-04 - val_loss: 5.2229e-04\n",
      "Epoch 9/25\n",
      "23/23 [==============================] - 22s 957ms/step - loss: 4.5543e-04 - val_loss: 5.2050e-04\n",
      "Epoch 10/25\n",
      "23/23 [==============================] - 22s 959ms/step - loss: 4.0730e-04 - val_loss: 5.0544e-04\n",
      "Epoch 11/25\n",
      "23/23 [==============================] - 22s 971ms/step - loss: 3.5668e-04 - val_loss: 5.0049e-04\n",
      "Epoch 12/25\n",
      "23/23 [==============================] - 22s 969ms/step - loss: 3.1238e-04 - val_loss: 4.8938e-04\n",
      "Epoch 13/25\n",
      "23/23 [==============================] - 22s 968ms/step - loss: 2.8244e-04 - val_loss: 4.7345e-04\n",
      "Epoch 14/25\n",
      "23/23 [==============================] - 22s 967ms/step - loss: 2.5848e-04 - val_loss: 4.9444e-04\n",
      "Epoch 15/25\n",
      "23/23 [==============================] - 22s 968ms/step - loss: 2.3532e-04 - val_loss: 4.7271e-04\n",
      "Epoch 16/25\n",
      "23/23 [==============================] - 22s 968ms/step - loss: 2.0822e-04 - val_loss: 4.7222e-04\n",
      "Epoch 17/25\n",
      "23/23 [==============================] - 22s 977ms/step - loss: 1.9010e-04 - val_loss: 4.7063e-04\n",
      "Epoch 18/25\n",
      "23/23 [==============================] - 23s 981ms/step - loss: 1.7218e-04 - val_loss: 4.7588e-04\n",
      "Epoch 19/25\n",
      "23/23 [==============================] - 23s 983ms/step - loss: 1.6276e-04 - val_loss: 4.6656e-04\n",
      "Epoch 20/25\n",
      "23/23 [==============================] - 23s 981ms/step - loss: 1.4449e-04 - val_loss: 4.7725e-04\n",
      "Epoch 21/25\n",
      "23/23 [==============================] - 23s 981ms/step - loss: 1.3107e-04 - val_loss: 4.6940e-04\n",
      "Epoch 22/25\n",
      "23/23 [==============================] - 23s 981ms/step - loss: 1.2107e-04 - val_loss: 4.6742e-04\n",
      "Epoch 23/25\n",
      "23/23 [==============================] - 23s 980ms/step - loss: 1.1141e-04 - val_loss: 4.6432e-04\n",
      "Epoch 24/25\n",
      "23/23 [==============================] - 23s 980ms/step - loss: 1.0185e-04 - val_loss: 4.6116e-04\n",
      "Epoch 25/25\n",
      "23/23 [==============================] - 23s 979ms/step - loss: 9.6006e-05 - val_loss: 4.7792e-04\n"
     ]
    }
   ],
   "source": [
    "H = model.fit(\n",
    "trainImages, trainTargets,\n",
    "validation_data=(testImages, testTargets),\n",
    "batch_size=32,\n",
    "epochs=25,\n",
    "verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save(\"localization.h5\", save_format=\"h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x26c9cc73248>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "N = H.params['epochs']\n",
    "plt.plot(np.arange(0, N), H.history[\"loss\"], label=\"train_loss\")\n",
    "plt.plot(np.arange(0, N), H.history[\"val_loss\"], label=\"val_loss\")\n",
    "plt.title(\"Bounding Box Regression Loss on Training Set\")\n",
    "plt.xlabel(\"Epoch #\")\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.legend(loc=\"lower left\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Source:\n",
    "\n",
    "https://www.pyimagesearch.com/2020/10/05/object-detection-bounding-box-regression-with-keras-tensorflow-and-deep-learning/"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tf2",
   "language": "python",
   "name": "tf2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}