{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "07bdc0da",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import json\n",
    "import numpy as np\n",
    "from tqdm.notebook import tqdm\n",
    "path_dataset = r'D:\\downloads\\Nurburg-karussel'\n",
    "from SuperviselyKeypointsGUI.SuperviselyKeypointsGUI import *\n",
    "\n",
    "\n",
    "keypoints_3d_path = r'SuperviselyKeypointsGUI\\karussel_24kps.csv'\n",
    "keypoints_3d = pd.read_csv(keypoints_3d_path, index_col=0).astype(float)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "ca890aa0",
   "metadata": {},
   "outputs": [],
   "source": [
    "def find_image(id):\n",
    "    for row in coco['images']:\n",
    "        if row['id'] == id:\n",
    "            return row"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "b8135031",
   "metadata": {},
   "outputs": [],
   "source": [
    "def ltrb_from_cloud(cloud_2d, imgSize, expansion=0.1):\n",
    "    height, width = imgSize\n",
    "    ltrb = np.round((cloud_2d[:, 0].min(), cloud_2d[:, 1].min(),\n",
    "                     cloud_2d[:, 0].max(), cloud_2d[:, 1].max())).astype(int)\n",
    "    \n",
    "    if expansion > 0:\n",
    "        dx = np.round((ltrb[2]-ltrb[0])*expansion/2)\n",
    "        dy = np.round((ltrb[3]-ltrb[1])*expansion/2)\n",
    "        ltrb += np.array([-dx, -2*dy, dx, dy], dtype=int)\n",
    "        \n",
    "    ltrb[[0,2]] = np.clip(ltrb[[0,2]], 0, width)\n",
    "    ltrb[[1,3]] = np.clip(ltrb[[1,3]], 0, height)\n",
    "    \n",
    "    return ltrb\n",
    "\n",
    "def ltrb2ltwh(ltrb):\n",
    "    return np.array([ltrb[0], ltrb[1], ltrb[2]-ltrb[0], ltrb[3]-ltrb[1]], dtype=int)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "79762c08",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(path_dataset, 'karusel_COCO.json'), 'r') as file:\n",
    "    coco = json.load(file)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a1cfe0e2",
   "metadata": {},
   "source": [
    "### Bboxes from pose"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f9440b54",
   "metadata": {},
   "outputs": [],
   "source": [
    "for obj in tqdm(coco['annotations']):\n",
    "    keypoints_2d = np.array(obj['keypoints']).reshape((-1, 3))[:, :2]\n",
    "    id, width, height, file_name = find_image(obj['image_id']).values()\n",
    "    \n",
    "    bbox_ltrb = ltrb_from_cloud(keypoints_2d, (height, width), 0.4)\n",
    "    bbox_ltwh = ltrb2ltwh(bbox_ltrb).tolist()\n",
    "    obj['bbox'] = bbox_ltwh\n",
    "    obj['area'] = bbox_ltwh[2]*bbox_ltwh[3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dfcb675b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# with open(os.path.join(path_dataset, 'karusel_COCO.json'), 'w') as file:\n",
    "#     json.dump(coco, file)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e8ff698d",
   "metadata": {},
   "source": [
    "### Split COCO json to train/val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3aec7a2b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "import copy\n",
    "\n",
    "def split_coco_json(coco, test_size=0.2, random_state=0):\n",
    "    \n",
    "    train_idx, test_idx = train_test_split([i['id'] for i in coco['images']],\n",
    "                                           test_size=test_size, random_state=random_state)\n",
    "\n",
    "\n",
    "    train = copy.deepcopy(coco)\n",
    "    test = copy.deepcopy(coco)\n",
    "\n",
    "    test['images'] = [x for x in coco['images'] if x['id'] in test_idx]\n",
    "    train['images'] = [x for x in coco['images'] if x['id'] in train_idx]\n",
    "\n",
    "    test['annotations'] = [x for x in coco['annotations'] if x['image_id'] in test_idx]\n",
    "    train['annotations'] = [x for x in coco['annotations'] if x['image_id'] in train_idx]\n",
    "    return train, test"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1cdd1a3c",
   "metadata": {},
   "source": [
    "### Create new splited dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aafd54fb",
   "metadata": {},
   "outputs": [],
   "source": [
    "train, test = split_coco_json(coco, 0.1, random_state=777)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f6af98fc",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_path_images = [os.path.join(path_dataset, 'images', 'img', x['file_name']) for x in test['images']]\n",
    "train_path_images = [os.path.join(path_dataset, 'images', 'img', x['file_name']) for x in train['images']]\n",
    "\n",
    "import shutil\n",
    "path_new_dataset = r'C:\\Users\\Kir\\Jupiter\\Nurburg\\OpenPifPaf\\Training\\Karusel_dataset'\n",
    "\n",
    "path_train_img = os.path.join(path_new_dataset, 'images', 'train')\n",
    "path_test_img = os.path.join(path_new_dataset, 'images', 'val')\n",
    "path_ann = os.path.join(path_new_dataset, 'annotations')\n",
    "\n",
    "os.makedirs(path_train_img, exist_ok=True)\n",
    "os.makedirs(path_test_img, exist_ok=True)\n",
    "os.makedirs(path_ann, exist_ok=True)\n",
    "\n",
    "with open(os.path.join(path_ann, 'train.json'), 'w') as file:\n",
    "    json.dump(train, file)\n",
    "    \n",
    "with open(os.path.join(path_ann, 'val.json'), 'w') as file:\n",
    "    json.dump(test, file)\n",
    "\n",
    "for path in train_path_images:\n",
    "    shutil.copy(path, path_train_img)\n",
    "\n",
    "for path in test_path_images:\n",
    "    shutil.copy(path, path_test_img)"
   ]
  }
 ],
 "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": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}