crop_and_augment.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "import os\n",
    "import itertools"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL.ImageOps import flip, mirror\n",
    "import PIL"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "343"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "folder_to_scan = './datasets/curated/'\n",
    "files = os.listdir(folder_to_scan)\n",
    "len(files)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "valid_file_types = ['.jpg']\n",
    "valid_files = []\n",
    "for file in files:\n",
    "    f, fe = os.path.splitext(file)\n",
    "    if fe in valid_file_types:\n",
    "        valid_files.append(file)\n",
    "full_path_valid_files = [folder_to_scan+file for file in valid_files]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_1d_limits(wind, limit, thresh):\n",
    "    x_left = []\n",
    "    x_right = []\n",
    "    if limit >= wind:\n",
    "        x_lim_reached = False\n",
    "        i = 0\n",
    "        while not x_lim_reached:\n",
    "            x_l = i * wind\n",
    "            x_r = (i + 1) * wind\n",
    "\n",
    "            if x_r <= limit:\n",
    "                x_right.append(x_r)\n",
    "                x_left.append(x_l)\n",
    "            else:\n",
    "                x_lim_reached = True\n",
    "                # some extra padding\n",
    "                if (x_r - limit) / wind < thresh:\n",
    "                    x_r = limit\n",
    "                    x_l = limit - wind\n",
    "                    x_right.append(x_r)\n",
    "                    x_left.append(x_l)\n",
    "            i += 1\n",
    "    return (x_left, x_right)\n",
    "\n",
    "\n",
    "def generate_cropping_boxes_from_limits(x_left, x_rigth, x_bottom, x_top):\n",
    "    croping_boxes = []\n",
    "    x_lims = [(x_l, x_r) for x_l, x_r in zip(x_left, x_rigth)]\n",
    "    y_lims = [(x_l, x_r) for x_l, x_r in zip(x_bottom, x_top)]\n",
    "    bounding_boxes = list(itertools.product(x_lims, y_lims))\n",
    "    for i in range(len(bounding_boxes)):\n",
    "        ((x1, x2), (y1, y2)) = bounding_boxes[i]\n",
    "        croping_boxes.append((x1, y1, x2, y2))\n",
    "    return croping_boxes\n",
    "\n",
    "\n",
    "def generate_cropping_boxes(image, cropping_window, tresh):\n",
    "    image_width, image_height = image.size\n",
    "    x_left, x_rigth = generate_1d_limits(cropping_window, image_width, tresh)\n",
    "    x_bottom, x_top = generate_1d_limits(cropping_window, image_height, tresh)\n",
    "    croping_boxes = generate_cropping_boxes_from_limits(x_left, x_rigth, x_bottom, x_top)\n",
    "    return croping_boxes\n",
    "\n",
    "\n",
    "def image_square_resize(im_input, new_size):\n",
    "    im = im_input.copy()\n",
    "    im = im.resize((new_size, new_size), PIL.Image.ANTIALIAS)    \n",
    "    return im\n",
    "\n",
    "\n",
    "def image_rotator(im_input, angle):\n",
    "    if angle==90:\n",
    "        return im_input.transpose(Image.ROTATE_90) \n",
    "    elif angle == 180:\n",
    "        return im_input.transpose(Image.ROTATE_180) \n",
    "    elif angle == 270:\n",
    "        return im_input.transpose(Image.ROTATE_270) \n",
    "    else:\n",
    "        raise ValueError('angle not supported')    \n",
    "        \n",
    "        \n",
    "def image_augmentator(im_input, return_orig = True):\n",
    "    im_aug = []\n",
    "    if return_orig:\n",
    "        im_aug.append(im_input.copy())\n",
    "    # 1.flip\n",
    "    im_aug.append(flip(im_input.copy()))\n",
    "    # 2. rot 180\n",
    "    im_aug.append(image_rotator(im_input.copy(), 180))\n",
    "    # 3. flip(rot_90)\n",
    "    im_aug.append(flip(image_rotator(im_input.copy(), 90)))\n",
    "    # 4. flip(rot_270)\n",
    "    im_aug.append(flip(image_rotator(im_input.copy(), 270)))\n",
    "    return im_aug        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# class SizeCounter:\n",
    "#     def __init__(self):\n",
    "#         self.counter = {}\n",
    "    \n",
    "#     def add(self, size):\n",
    "#         try:\n",
    "#             self.counter[size]\n",
    "#             self.counter[size] +=1\n",
    "#         except KeyError:\n",
    "#             self.counter[size] = 1\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# sizes = SizeCounter()\n",
    "# for file in full_path_valid_files:\n",
    "#     im = Image.open(file)\n",
    "#     sizes.add(size=im.size)\n",
    "    \n",
    "# # sizes.counter"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### No augmentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "cropping_window = 512\n",
    "saving_folder_name = './datasets/cropped_files_no_aug/'\n",
    "padding_tresh = 0.25\n",
    "resize = False\n",
    "image_output_size = 512\n",
    "\n",
    "os.makedirs(saving_folder_name, exist_ok=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4120\n"
     ]
    }
   ],
   "source": [
    "for file in full_path_valid_files:\n",
    "    im = Image.open(file)\n",
    "    croping_boxes = generate_cropping_boxes(im, cropping_window, padding_tresh) \n",
    "    file_name = os.path.basename(file)\n",
    "    base_name, file_ext = os.path.splitext(file_name)\n",
    "    for i,b in enumerate(croping_boxes):\n",
    "        imc = im.crop(b) #left bottom right upper\n",
    "        if resize: \n",
    "            imc = image_square_resize(imc,image_output_size )\n",
    "        f_name = saving_folder_name + base_name +'_{}_'.format(i) + file_ext\n",
    "        imc.save(f_name)\n",
    "print(len(os.listdir(saving_folder_name)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Aumentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "cropping_window = 512\n",
    "saving_folder_name = './datasets/cropped_files_aug/'\n",
    "padding_tresh = 0.25\n",
    "resize = False\n",
    "image_output_size = 512\n",
    "\n",
    "os.makedirs(saving_folder_name, exist_ok=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "20600\n"
     ]
    }
   ],
   "source": [
    "for file in full_path_valid_files:    \n",
    "    im = Image.open(file)\n",
    "    for ia, im in enumerate(image_augmentator(im)):\n",
    "        croping_boxes = generate_cropping_boxes(im, cropping_window, padding_tresh) \n",
    "        file_name = os.path.basename(file)\n",
    "        base_name, file_ext = os.path.splitext(file_name)\n",
    "        for i,b in enumerate(croping_boxes):\n",
    "            imc = im.crop(b) #left bottom right upper\n",
    "            if resize: \n",
    "                imc = image_square_resize(imc,image_output_size )\n",
    "            f_name = saving_folder_name + base_name +'_{}_{}_'.format(ia,i) + file_ext\n",
    "            imc.save(f_name)\n",
    "    \n",
    "print(len(os.listdir(saving_folder_name)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "testing-kernel",
   "language": "python",
   "name": "testing-kernel"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}