getdata.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ac5534c4",
   "metadata": {},
   "outputs": [],
   "source": [
    "!unzip data.zip\n",
    "!wget https://storage.googleapis.com/sos-aiplayground-operating-environment/data.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "9c4a3835-ec83-449a-935a-2766e73d1205",
   "metadata": {},
   "outputs": [],
   "source": [
    "import glob\n",
    "import os\n",
    "import numpy as np\n",
    "import cv2\n",
    "import albumentations as albu\n",
    "from PIL import Image\n",
    "import shutil\n",
    "from torchvision.transforms import (Pad, ColorJitter, Resize, FiveCrop, RandomCrop,\n",
    "                                    RandomHorizontalFlip, RandomRotation, RandomVerticalFlip)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "940d7f6f-8ac6-450b-ab33-28aad326e561",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = [\"top_mosaic_09cm_area1\",\n",
    "             \"top_mosaic_09cm_area3\",\n",
    "             \"top_mosaic_09cm_area5\",\n",
    "             \"top_mosaic_09cm_area7\",\n",
    "             \"top_mosaic_09cm_area11\",\n",
    "             \"top_mosaic_09cm_area13\",\n",
    "             \"top_mosaic_09cm_area15\",\n",
    "             \"top_mosaic_09cm_area17\",\n",
    "             \"top_mosaic_09cm_area21\",\n",
    "             \"top_mosaic_09cm_area23\",\n",
    "             \"top_mosaic_09cm_area26\",\n",
    "             \"top_mosaic_09cm_area28\",\n",
    "             \"top_mosaic_09cm_area30\",\n",
    "             \"top_mosaic_09cm_area32\",\n",
    "             \"top_mosaic_09cm_area34\",\n",
    "             \"top_mosaic_09cm_area37\"]\n",
    "train_imgs_output_dir = \"./data/train/images\"\n",
    "train_masks_output_dir = \"./data/train/masks\"\n",
    "val_imgs_output_dir = \"./data/val/images\"\n",
    "val_masks_output_dir = \"./data/val/masks\"\n",
    "test_imgs_output_dir = \"./data/test/images\"\n",
    "test_masks_output_dir = \"./data/test/masks\"\n",
    "if not os.path.exists(train_imgs_output_dir):\n",
    "    os.makedirs(train_imgs_output_dir)\n",
    "if not os.path.exists(train_masks_output_dir):\n",
    "    os.makedirs(train_masks_output_dir)\n",
    "if not os.path.exists(val_imgs_output_dir):\n",
    "    os.makedirs(val_imgs_output_dir)\n",
    "if not os.path.exists(val_masks_output_dir):\n",
    "    os.makedirs(val_masks_output_dir)\n",
    "if not os.path.exists(test_imgs_output_dir):\n",
    "    os.makedirs(test_imgs_output_dir)\n",
    "if not os.path.exists(test_masks_output_dir):\n",
    "    os.makedirs(test_masks_output_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "2ebe443b-e28b-4a66-937a-3b33387f4ce5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['./data/images/top_mosaic_09cm_area29.tif', './data/images/top_mosaic_09cm_area15.tif', './data/images/top_mosaic_09cm_area14.tif', './data/images/top_mosaic_09cm_area28.tif', './data/images/top_mosaic_09cm_area16.tif', './data/images/top_mosaic_09cm_area17.tif', './data/images/top_mosaic_09cm_area13.tif', './data/images/top_mosaic_09cm_area12.tif', './data/images/top_mosaic_09cm_area10.tif', './data/images/top_mosaic_09cm_area38.tif', './data/images/top_mosaic_09cm_area11.tif', './data/images/top_mosaic_09cm_area5.tif', './data/images/top_mosaic_09cm_area4.tif', './data/images/top_mosaic_09cm_area6.tif', './data/images/top_mosaic_09cm_area7.tif', './data/images/top_mosaic_09cm_area3.tif', './data/images/top_mosaic_09cm_area2.tif', './data/images/top_mosaic_09cm_area1.tif', './data/images/top_mosaic_09cm_area8.tif', './data/images/top_mosaic_09cm_area20.tif', './data/images/top_mosaic_09cm_area34.tif', './data/images/top_mosaic_09cm_area35.tif', './data/images/top_mosaic_09cm_area21.tif', './data/images/top_mosaic_09cm_area37.tif', './data/images/top_mosaic_09cm_area23.tif', './data/images/top_mosaic_09cm_area22.tif', './data/images/top_mosaic_09cm_area32.tif', './data/images/top_mosaic_09cm_area26.tif', './data/images/top_mosaic_09cm_area27.tif', './data/images/top_mosaic_09cm_area33.tif', './data/images/top_mosaic_09cm_area31.tif', './data/images/top_mosaic_09cm_area30.tif', './data/images/top_mosaic_09cm_area24.tif']\n",
      "['./data/masks/top_mosaic_09cm_area29.tif', './data/masks/top_mosaic_09cm_area15.tif', './data/masks/top_mosaic_09cm_area14.tif', './data/masks/top_mosaic_09cm_area28.tif', './data/masks/top_mosaic_09cm_area16.tif', './data/masks/top_mosaic_09cm_area17.tif', './data/masks/top_mosaic_09cm_area13.tif', './data/masks/top_mosaic_09cm_area12.tif', './data/masks/top_mosaic_09cm_area10.tif', './data/masks/top_mosaic_09cm_area38.tif', './data/masks/top_mosaic_09cm_area11.tif', './data/masks/top_mosaic_09cm_area5.tif', './data/masks/top_mosaic_09cm_area4.tif', './data/masks/top_mosaic_09cm_area6.tif', './data/masks/top_mosaic_09cm_area7.tif', './data/masks/top_mosaic_09cm_area3.tif', './data/masks/top_mosaic_09cm_area2.tif', './data/masks/top_mosaic_09cm_area1.tif', './data/masks/top_mosaic_09cm_area8.tif', './data/masks/top_mosaic_09cm_area20.tif', './data/masks/top_mosaic_09cm_area34.tif', './data/masks/top_mosaic_09cm_area35.tif', './data/masks/top_mosaic_09cm_area21.tif', './data/masks/top_mosaic_09cm_area37.tif', './data/masks/top_mosaic_09cm_area23.tif', './data/masks/top_mosaic_09cm_area22.tif', './data/masks/top_mosaic_09cm_area32.tif', './data/masks/top_mosaic_09cm_area26.tif', './data/masks/top_mosaic_09cm_area27.tif', './data/masks/top_mosaic_09cm_area33.tif', './data/masks/top_mosaic_09cm_area31.tif', './data/masks/top_mosaic_09cm_area30.tif', './data/masks/top_mosaic_09cm_area24.tif']\n",
      "['./data/boundarys/top_mosaic_09cm_area38_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area30_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area37_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area22_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area23_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area31_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area24_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area33_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area26_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area21_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area34_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area29_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area28_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area35_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area20_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area27_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area32_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area7_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area13_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area8_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area14_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area1_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area15_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area9_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area12_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area6_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area4_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area10_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area17_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area3_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area2_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area16_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area11_noBoundary.tif', './data/boundarys/top_mosaic_09cm_area5_noBoundary.tif']\n"
     ]
    }
   ],
   "source": [
    "imgs_dir=\"./data/images\"\n",
    "masks_dir=\"./data/masks\"\n",
    "boundarys_dir=\"./data/boundarys\"\n",
    "img_paths = glob.glob(os.path.join(imgs_dir, \"*.tif\"))\n",
    "mask_paths = glob.glob(os.path.join(masks_dir, \"*.tif\"))\n",
    "boundary_paths = glob.glob(os.path.join(boundarys_dir, \"*.tif\"))\n",
    "print(img_paths)\n",
    "print(mask_paths)\n",
    "print(boundary_paths)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "1d1d3180-ed64-477c-adf5-393ff170a523",
   "metadata": {},
   "outputs": [],
   "source": [
    "ImSurf = np.array([255, 255, 255])  # label 0\n",
    "Building = np.array([255, 0, 0]) # label 1\n",
    "LowVeg = np.array([255, 255, 0]) # label 2\n",
    "Tree = np.array([0, 255, 0]) # label 3\n",
    "Car = np.array([0, 255, 255]) # label 4\n",
    "Clutter = np.array([0, 0, 255]) # label 5\n",
    "Boundary = np.array([0, 0, 0]) # label 6\n",
    "num_classes = 6\n",
    "\n",
    "def get_img_mask_padded(image, mask, patch_size, mode):\n",
    "    img, mask = np.array(image), np.array(mask)\n",
    "    oh, ow = img.shape[0], img.shape[1]\n",
    "    rh, rw = oh % patch_size, ow % patch_size\n",
    "    width_pad = 0 if rw == 0 else patch_size - rw\n",
    "    height_pad = 0 if rh == 0 else patch_size - rh\n",
    "\n",
    "    h, w = oh + height_pad, ow + width_pad\n",
    "    pad_img = albu.PadIfNeeded(min_height=h, min_width=w, position='bottom_right',\n",
    "                               border_mode=0, value=[0, 0, 0])(image=img)\n",
    "    if mode == 'train':\n",
    "        pad_img = albu.PadIfNeeded(min_height=h, min_width=w, position='bottom_right')(image=img)\n",
    "\n",
    "    pad_mask = albu.PadIfNeeded(min_height=h, min_width=w, position='bottom_right',\n",
    "                                border_mode=0, value=[0, 0, 0])(image=mask)\n",
    "    img_pad, mask_pad = pad_img['image'], pad_mask['image']\n",
    "    img_pad = cv2.cvtColor(np.array(img_pad), cv2.COLOR_RGB2BGR)\n",
    "    mask_pad = cv2.cvtColor(np.array(mask_pad), cv2.COLOR_RGB2BGR)\n",
    "    return img_pad, mask_pad\n",
    "\n",
    "\n",
    "def pv2rgb(mask):\n",
    "    h, w = mask.shape[0], mask.shape[1]\n",
    "    mask_rgb = np.zeros(shape=(h, w, 3), dtype=np.uint8)\n",
    "    mask_convert = mask[np.newaxis, :, :]\n",
    "    mask_rgb[np.all(mask_convert == 3, axis=0)] = [0, 255, 0]\n",
    "    mask_rgb[np.all(mask_convert == 0, axis=0)] = [255, 255, 255]\n",
    "    mask_rgb[np.all(mask_convert == 1, axis=0)] = [255, 0, 0]\n",
    "    mask_rgb[np.all(mask_convert == 2, axis=0)] = [255, 255, 0]\n",
    "    mask_rgb[np.all(mask_convert == 4, axis=0)] = [0, 204, 255]\n",
    "    mask_rgb[np.all(mask_convert == 5, axis=0)] = [0, 0, 255]\n",
    "    return mask_rgb\n",
    "\n",
    "\n",
    "def car_color_replace(mask):\n",
    "    mask = cv2.cvtColor(np.array(mask.copy()), cv2.COLOR_RGB2BGR)\n",
    "    mask[np.all(mask == [0, 255, 255], axis=-1)] = [0, 204, 255]\n",
    "\n",
    "    return mask\n",
    "\n",
    "\n",
    "def rgb_to_2D_label(_label):\n",
    "    print(_label.size)\n",
    "    _label = _label.transpose(2, 0, 1)\n",
    "    label_seg = np.zeros(_label.shape[1:], dtype=np.uint8)\n",
    "    label_seg[np.all(_label.transpose([1, 2, 0]) == ImSurf, axis=-1)] = 0\n",
    "    label_seg[np.all(_label.transpose([1, 2, 0]) == Building, axis=-1)] = 1\n",
    "    label_seg[np.all(_label.transpose([1, 2, 0]) == LowVeg, axis=-1)] = 2\n",
    "    label_seg[np.all(_label.transpose([1, 2, 0]) == Tree, axis=-1)] = 3\n",
    "    label_seg[np.all(_label.transpose([1, 2, 0]) == Car, axis=-1)] = 4\n",
    "    label_seg[np.all(_label.transpose([1, 2, 0]) == Clutter, axis=-1)] = 5\n",
    "    # label_seg[np.all(_label.transpose([1, 2, 0]) == Boundary, axis=-1)] = 6\n",
    "    return label_seg\n",
    "\n",
    "\n",
    "def image_augment(image, mask, patch_size, mode='train', val_scale=1.0):\n",
    "    image_list = []\n",
    "    mask_list = []\n",
    "    image_width, image_height = image.size[1], image.size[0]\n",
    "    mask_width, mask_height = mask.size[1], mask.size[0]\n",
    "    assert image_height == mask_height and image_width == mask_width\n",
    "    if mode == 'train':\n",
    "        # resize_0 = Resize(size=(int(image_width * 0.25), int(image_height * 0.25)))\n",
    "        # resize_1 = Resize(size=(int(image_width * 0.5), int(image_height * 0.5)))\n",
    "        # resize_2 = Resize(size=(int(image_width * 0.75), int(image_height * 0.75)))\n",
    "        # resize_3 = Resize(size=(int(image_width * 1.25), int(image_height * 1.25)))\n",
    "        # resize_4 = Resize(size=(int(image_width * 1.5), int(image_height * 1.5)))\n",
    "        # resize_5 = Resize(size=(int(image_width * 1.75), int(image_height * 1.75)))\n",
    "        # resize_6 = Resize(size=(int(image_width * 2.0), int(image_height * 2.0)))\n",
    "        # image_resize_0, mask_resize_0 = resize_0(image.copy()), resize_0(mask.copy())\n",
    "        # image_resize_1, mask_resize_1 = resize_1(image.copy()), resize_1(mask.copy())\n",
    "        # image_resize_2, mask_resize_2 = resize_2(image.copy()), resize_2(mask.copy())\n",
    "        # image_resize_3, mask_resize_3 = resize_3(image.copy()), resize_3(mask.copy())\n",
    "        # image_resize_4, mask_resize_4 = resize_4(image.copy()), resize_4(mask.copy())\n",
    "        # image_resize_5, mask_resize_5 = resize_5(image.copy()), resize_5(mask.copy())\n",
    "        # image_resize_6, mask_resize_6 = resize_6(image.copy()), resize_6(mask.copy())\n",
    "        h_vlip = RandomHorizontalFlip(p=1.0)\n",
    "        v_vlip = RandomVerticalFlip(p=1.0)\n",
    "        # crop_1 = RandomCrop(size=(int(image_width*0.75), int(image_height*0.75)))\n",
    "        # crop_2 = RandomCrop(size=(int(image_width * 0.5), int(image_height * 0.5)))\n",
    "        # color = torchvision.transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2)\n",
    "        image_h_vlip, mask_h_vlip = h_vlip(image.copy()), h_vlip(mask.copy())\n",
    "        image_v_vlip, mask_v_vlip = v_vlip(image.copy()), v_vlip(mask.copy())\n",
    "        # image_crop_1, mask_crop_1 = crop_1(image.copy()), crop_1(mask.copy())\n",
    "        # image_crop_2, mask_crop_2 = crop_2(image.copy()), crop_2(mask.copy())\n",
    "        # image_color = color(image.copy())\n",
    "\n",
    "        image_list_train = [image, image_h_vlip, image_v_vlip]\n",
    "        mask_list_train = [mask, mask_h_vlip, mask_v_vlip]\n",
    "        # image_list_train = [image]\n",
    "        # mask_list_train = [mask]\n",
    "        for i in range(len(image_list_train)):\n",
    "            image_tmp, mask_tmp = get_img_mask_padded(image_list_train[i], mask_list_train[i], patch_size, mode)\n",
    "            mask_tmp = rgb_to_2D_label(mask_tmp.copy())\n",
    "            image_list.append(image_tmp)\n",
    "            mask_list.append(mask_tmp)\n",
    "    else:\n",
    "        rescale = Resize(size=(int(image_width * val_scale), int(image_height * val_scale)))\n",
    "        image, mask = rescale(image.copy()), rescale(mask.copy())\n",
    "        image, mask = get_img_mask_padded(image.copy(), mask.copy(), patch_size, mode)\n",
    "        mask = rgb_to_2D_label(mask.copy())\n",
    "        image_list.append(image)\n",
    "        mask_list.append(mask)\n",
    "    return image_list, mask_list\n",
    "\n",
    "\n",
    "def randomsizedcrop(image, mask):\n",
    "    # assert image.shape[:2] == mask.shape\n",
    "    h, w = image.shape[0], image.shape[1]\n",
    "    crop = albu.RandomSizedCrop(min_max_height=(int(3*h//8), int(h//2)), width=h, height=w)(image=image.copy(), mask=mask.copy())\n",
    "    img_crop, mask_crop = crop['image'], crop['mask']\n",
    "    return img_crop, mask_crop\n",
    "\n",
    "\n",
    "def car_aug(image, mask):\n",
    "    assert image.shape[:2] == mask.shape\n",
    "    v_flip = albu.VerticalFlip(p=1.0)(image=image.copy(), mask=mask.copy())\n",
    "    h_flip = albu.HorizontalFlip(p=1.0)(image=image.copy(), mask=mask.copy())\n",
    "    rotate_90 = albu.RandomRotate90(p=1.0)(image=image.copy(), mask=mask.copy())\n",
    "    # blur = albu.GaussianBlur(p=1.0)(image=image.copy())\n",
    "    image_vflip, mask_vflip = v_flip['image'], v_flip['mask']\n",
    "    image_hflip, mask_hflip = h_flip['image'], h_flip['mask']\n",
    "    image_rotate, mask_rotate = rotate_90['image'], rotate_90['mask']\n",
    "    # blur_image = blur['image']\n",
    "    image_list = [image, image_vflip, image_hflip, image_rotate]\n",
    "    mask_list = [mask, mask_vflip, mask_hflip, mask_rotate]\n",
    "\n",
    "    return image_list, mask_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "8b26bba3-ff05-42d8-ab13-c65379313f9d",
   "metadata": {},
   "outputs": [],
   "source": [
    "def vaihingen_format(inp):\n",
    "    (img_path, mask_path, imgs_output_dir, masks_output_dir, eroded, gt, mode, val_scale, split_size, stride) = inp\n",
    "    img_filename = os.path.splitext(os.path.basename(img_path))[0]\n",
    "    mask_filename = os.path.splitext(os.path.basename(mask_path))[0]\n",
    "    if eroded:\n",
    "        mask_path = mask_path[:-4] + '_noBoundary.tif'\n",
    "    img = Image.open(img_path).convert('RGB')\n",
    "    mask = Image.open(mask_path).convert('RGB')\n",
    "    \n",
    "    image_pad, mask_pad = get_img_mask_padded(img.copy(), mask.copy(), 16, \"train\")\n",
    "    mask_tmp = rgb_to_2D_label(mask_pad.copy())\n",
    "    out_mask_path = os.path.join(test_masks_output_dir,\n",
    "                                 \"{}.png\".format(mask_filename))\n",
    "    cv2.imwrite(out_mask_path, mask_tmp)\n",
    "    out_img_path = os.path.join(test_imgs_output_dir,\n",
    "                                \"{}.tif\".format(img_filename))\n",
    "\n",
    "    cv2.imwrite(out_img_path, image_pad)\n",
    "    if gt:\n",
    "        mask_ = car_color_replace(mask)\n",
    "        out_origin_mask_path = os.path.join(masks_output_dir + '/origin/', \"{}.tif\".format(mask_filename))\n",
    "        cv2.imwrite(out_origin_mask_path, mask_)\n",
    "    # print(img_path)\n",
    "    # print(img.size, mask.size)\n",
    "    # img and mask shape: WxHxC\n",
    "    image_list, mask_list = image_augment(image=img.copy(), mask=mask.copy(), patch_size=split_size,\n",
    "                                          mode=mode, val_scale=val_scale)\n",
    "    assert img_filename == mask_filename and len(image_list) == len(mask_list)\n",
    "    for m in range(len(image_list)):\n",
    "        k = 0\n",
    "        img = image_list[m]\n",
    "        mask = mask_list[m]\n",
    "        assert img.shape[0] == mask.shape[0] and img.shape[1] == mask.shape[1]\n",
    "        if gt:\n",
    "            mask = pv2rgb(mask)\n",
    "\n",
    "        for y in range(0, img.shape[0], stride):\n",
    "            for x in range(0, img.shape[1], stride):\n",
    "                img_tile = img[y:y + split_size, x:x + split_size]\n",
    "                mask_tile = mask[y:y + split_size, x:x + split_size]\n",
    "\n",
    "                if img_tile.shape[0] == split_size and img_tile.shape[1] == split_size \\\n",
    "                        and mask_tile.shape[0] == split_size and mask_tile.shape[1] == split_size:\n",
    "                    image_crop, mask_crop = randomsizedcrop(img_tile, mask_tile)\n",
    "                    bins = np.array(range(num_classes + 1))\n",
    "                    class_pixel_counts, _ = np.histogram(mask_crop, bins=bins)\n",
    "                    cf = class_pixel_counts / (mask_crop.shape[0] * mask_crop.shape[1])\n",
    "                    if cf[4] > 0.1 and mode == 'train':\n",
    "                        car_imgs, car_masks = car_aug(image_crop, mask_crop)\n",
    "                        for i in range(len(car_imgs)):\n",
    "                            out_img_path = os.path.join(imgs_output_dir,\n",
    "                                                        \"{}_{}_{}_{}.tif\".format(img_filename, m, k, i))\n",
    "                            \n",
    "                            cv2.imwrite(out_img_path, car_imgs[i])\n",
    "\n",
    "                            out_mask_path = os.path.join(masks_output_dir,\n",
    "                                                         \"{}_{}_{}_{}.png\".format(mask_filename, m, k, i))\n",
    "                            cv2.imwrite(out_mask_path, car_masks[i])\n",
    "                    else:\n",
    "                        out_img_path = os.path.join(imgs_output_dir, \"{}_{}_{}.tif\".format(img_filename, m, k))\n",
    "                        \n",
    "                        cv2.imwrite(out_img_path, img_tile)\n",
    "\n",
    "                        out_mask_path = os.path.join(masks_output_dir, \"{}_{}_{}.png\".format(mask_filename, m, k))\n",
    "                        cv2.imwrite(out_mask_path, mask_tile)\n",
    "\n",
    "                k += 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "a86bbb0a-fcfb-407a-857f-5743f5325c67",
   "metadata": {},
   "outputs": [],
   "source": [
    "# imgs_output_dir = \"./data/train/images\"\n",
    "# masks_output_dir = \"./data/train/masks\"\n",
    "# if not os.path.exists(imgs_output_dir):\n",
    "#     os.makedirs(imgs_output_dir)\n",
    "# if not os.path.exists(masks_output_dir):\n",
    "#     os.makedirs(masks_output_dir)\n",
    "# for img_path, mask_path in zip(img_paths, mask_paths):\n",
    "#     print(img_path, mask_path, boundary_path)\n",
    "#     img_filename = os.path.splitext(os.path.basename(img_path))[0]\n",
    "#     mask_filename = os.path.splitext(os.path.basename(mask_path))[0]\n",
    "#     # mask_path = mask_path[:-4] + '_noBoundary.tif'\n",
    "#     img = Image.open(img_path).convert('RGB').resize((1900,2500))\n",
    "#     mask = Image.open(mask_path).convert('RGB').resize((1900,2500))\n",
    "#     print(img.size)\n",
    "#     mask = rgb_to_2D_label(mask)\n",
    "#     out_img_path = os.path.join(imgs_output_dir,img_filename)\n",
    "\n",
    "#     cv2.imwrite(out_img_path, car_imgs[i])\n",
    "\n",
    "#     out_mask_path = os.path.join(masks_output_dir,mask_filename)\n",
    "#     cv2.imwrite(out_mask_path, car_masks[i])\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "835394a1-daa4-4bd1-a482-07e688a90c86",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "./data/images/top_mosaic_09cm_area29.tif ./data/masks/top_mosaic_09cm_area29.tif\n",
      "top_mosaic_09cm_area29\n",
      "14837760\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area15.tif ./data/masks/top_mosaic_09cm_area15.tif\n",
      "top_mosaic_09cm_area15\n",
      "14837760\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area14.tif ./data/masks/top_mosaic_09cm_area14.tif\n",
      "top_mosaic_09cm_area14\n",
      "14837760\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area28.tif ./data/masks/top_mosaic_09cm_area28.tif\n",
      "top_mosaic_09cm_area28\n",
      "14837760\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area16.tif ./data/masks/top_mosaic_09cm_area16.tif\n",
      "top_mosaic_09cm_area16\n",
      "14837760\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area17.tif ./data/masks/top_mosaic_09cm_area17.tif\n",
      "top_mosaic_09cm_area17\n",
      "9082368\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area13.tif ./data/masks/top_mosaic_09cm_area13.tif\n",
      "top_mosaic_09cm_area13\n",
      "21749760\n",
      "28311552\n",
      "28311552\n",
      "28311552\n",
      "./data/images/top_mosaic_09cm_area12.tif ./data/masks/top_mosaic_09cm_area12.tif\n",
      "top_mosaic_09cm_area12\n",
      "14961408\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area10.tif ./data/masks/top_mosaic_09cm_area10.tif\n",
      "top_mosaic_09cm_area10\n",
      "14499840\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area38.tif ./data/masks/top_mosaic_09cm_area38.tif\n",
      "top_mosaic_09cm_area38\n",
      "29368320\n",
      "37748736\n",
      "./data/images/top_mosaic_09cm_area11.tif ./data/masks/top_mosaic_09cm_area11.tif\n",
      "top_mosaic_09cm_area11\n",
      "14714112\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area5.tif ./data/masks/top_mosaic_09cm_area5.tif\n",
      "top_mosaic_09cm_area5\n",
      "14499840\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area4.tif ./data/masks/top_mosaic_09cm_area4.tif\n",
      "top_mosaic_09cm_area4\n",
      "14499840\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area6.tif ./data/masks/top_mosaic_09cm_area6.tif\n",
      "top_mosaic_09cm_area6\n",
      "14499840\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area7.tif ./data/masks/top_mosaic_09cm_area7.tif\n",
      "top_mosaic_09cm_area7\n",
      "14499840\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area3.tif ./data/masks/top_mosaic_09cm_area3.tif\n",
      "top_mosaic_09cm_area3\n",
      "18192384\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area2.tif ./data/masks/top_mosaic_09cm_area2.tif\n",
      "top_mosaic_09cm_area2\n",
      "20195328\n",
      "28311552\n",
      "./data/images/top_mosaic_09cm_area1.tif ./data/masks/top_mosaic_09cm_area1.tif\n",
      "top_mosaic_09cm_area1\n",
      "14837760\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area8.tif ./data/masks/top_mosaic_09cm_area8.tif\n",
      "top_mosaic_09cm_area8\n",
      "14499840\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area20.tif ./data/masks/top_mosaic_09cm_area20.tif\n",
      "top_mosaic_09cm_area20\n",
      "13029120\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area34.tif ./data/masks/top_mosaic_09cm_area34.tif\n",
      "top_mosaic_09cm_area34\n",
      "10690560\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area35.tif ./data/masks/top_mosaic_09cm_area35.tif\n",
      "top_mosaic_09cm_area35\n",
      "15949824\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area21.tif ./data/masks/top_mosaic_09cm_area21.tif\n",
      "top_mosaic_09cm_area21\n",
      "14622720\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area37.tif ./data/masks/top_mosaic_09cm_area37.tif\n",
      "top_mosaic_09cm_area37\n",
      "12000000\n",
      "12582912\n",
      "12582912\n",
      "12582912\n",
      "./data/images/top_mosaic_09cm_area23.tif ./data/masks/top_mosaic_09cm_area23.tif\n",
      "top_mosaic_09cm_area23\n",
      "14622720\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area22.tif ./data/masks/top_mosaic_09cm_area22.tif\n",
      "top_mosaic_09cm_area22\n",
      "14622720\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area32.tif ./data/masks/top_mosaic_09cm_area32.tif\n",
      "top_mosaic_09cm_area32\n",
      "15237120\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area26.tif ./data/masks/top_mosaic_09cm_area26.tif\n",
      "top_mosaic_09cm_area26\n",
      "16171008\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area27.tif ./data/masks/top_mosaic_09cm_area27.tif\n",
      "top_mosaic_09cm_area27\n",
      "19169280\n",
      "25165824\n",
      "./data/images/top_mosaic_09cm_area33.tif ./data/masks/top_mosaic_09cm_area33.tif\n",
      "top_mosaic_09cm_area33\n",
      "12165120\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area31.tif ./data/masks/top_mosaic_09cm_area31.tif\n",
      "top_mosaic_09cm_area31\n",
      "15237120\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area30.tif ./data/masks/top_mosaic_09cm_area30.tif\n",
      "top_mosaic_09cm_area30\n",
      "14961408\n",
      "18874368\n",
      "18874368\n",
      "18874368\n",
      "./data/images/top_mosaic_09cm_area24.tif ./data/masks/top_mosaic_09cm_area24.tif\n",
      "top_mosaic_09cm_area24\n",
      "14622720\n",
      "18874368\n"
     ]
    }
   ],
   "source": [
    "\n",
    "for img_path, mask_path in zip(img_paths, mask_paths):\n",
    "    print(img_path, mask_path)\n",
    "    img_filename = os.path.splitext(os.path.basename(img_path))[0]\n",
    "    print(img_filename)\n",
    "    mask_filename = os.path.splitext(os.path.basename(mask_path))[0]\n",
    "    # mask_path = mask_path[:-4] + '_noBoundary.tif'\n",
    "    # img = Image.open(img_path).convert('RGB').resize((1900,2500))\n",
    "    # mask = Image.open(mask_path).convert('RGB').resize((1900,2500))\n",
    "    # print(img.size)\n",
    "    # img.show()\n",
    "    # mask = rgb_to_2D_label(mask.copy())\n",
    "    if img_filename in train_data:\n",
    "        inp = (img_path, mask_path, train_imgs_output_dir, train_masks_output_dir, False, False, \"train\", 1.0, 1024, 512)\n",
    "        vaihingen_format(inp)\n",
    "    else:\n",
    "        inp = (img_path, mask_path, val_imgs_output_dir, val_masks_output_dir, False, False, \"val\", 1.0, 1024, 1024)\n",
    "        vaihingen_format(inp)\n",
    "        \n",
    "    # else:\n",
    "    #     inp = (img_path, mask_path, test_imgs_output_dir, test_masks_output_dir, False, False, \"test\", 1.0, 1024, 1024)\n",
    "    #     vaihingen_format(inp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "37c37da9-257b-4ba8-a6bf-2a75ff0249bf",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/workspace/deeplabv3\n",
      "pretrained resnet, 50\n",
      "num_val_batches: 33\n",
      "/opt/conda/lib/python3.7/site-packages/torch/nn/functional.py:3722: UserWarning: nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.\n",
      "  warnings.warn(\"nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.\")\n",
      "train: {'mIoU': 0.7841946516638807, 'F1': 0.8766468598674816, 'OA': 0.8973981116564371}\n",
      "{'ImSurf': 0.8386540711186123, 'Building': 0.896125429020775, 'LowVeg': 0.7238784877630702, 'Tree': 0.7981346637017682, 'Car': 0.6641806067151775, 'Clutter': 0.5456420969100645}\n",
      "val loss: 0.330717\n"
     ]
    }
   ],
   "source": [
    "!python evaluation/eval_on_test.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "14e85375-f2a5-40d0-a36c-73dc157125e7",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.9.6 64-bit",
   "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": "3.9.6"
  },
  "vscode": {
   "interpreter": {
    "hash": "31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}