{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "b4f4da53",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pandas as pd\n",
    "from tqdm import tqdm\n",
    "from PIL import Image as Im\n",
    "from huggingface_hub import notebook_login\n",
    "from datasets import load_dataset, Dataset, Image"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aa9019fe",
   "metadata": {},
   "source": [
    "### Load a set with coordinates"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "11f31770",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>filename</th>\n",
       "      <th>x_from</th>\n",
       "      <th>y_from</th>\n",
       "      <th>width</th>\n",
       "      <th>height</th>\n",
       "      <th>sign_class</th>\n",
       "      <th>sign_id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>autosave01_02_2012_09_13_33.jpg</td>\n",
       "      <td>649</td>\n",
       "      <td>376</td>\n",
       "      <td>18</td>\n",
       "      <td>18</td>\n",
       "      <td>2_1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>autosave01_02_2012_09_13_34.jpg</td>\n",
       "      <td>671</td>\n",
       "      <td>356</td>\n",
       "      <td>20</td>\n",
       "      <td>21</td>\n",
       "      <td>2_1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>autosave01_02_2012_09_13_35.jpg</td>\n",
       "      <td>711</td>\n",
       "      <td>332</td>\n",
       "      <td>27</td>\n",
       "      <td>26</td>\n",
       "      <td>2_1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>autosave01_02_2012_09_13_36.jpg</td>\n",
       "      <td>764</td>\n",
       "      <td>290</td>\n",
       "      <td>37</td>\n",
       "      <td>36</td>\n",
       "      <td>2_1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>autosave01_02_2012_09_13_36.jpg</td>\n",
       "      <td>684</td>\n",
       "      <td>384</td>\n",
       "      <td>17</td>\n",
       "      <td>17</td>\n",
       "      <td>1_23</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                          filename  x_from  y_from  width  height sign_class  \\\n",
       "0  autosave01_02_2012_09_13_33.jpg     649     376     18      18        2_1   \n",
       "1  autosave01_02_2012_09_13_34.jpg     671     356     20      21        2_1   \n",
       "2  autosave01_02_2012_09_13_35.jpg     711     332     27      26        2_1   \n",
       "3  autosave01_02_2012_09_13_36.jpg     764     290     37      36        2_1   \n",
       "4  autosave01_02_2012_09_13_36.jpg     684     384     17      17       1_23   \n",
       "\n",
       "   sign_id  \n",
       "0        0  \n",
       "1        0  \n",
       "2        0  \n",
       "3        0  \n",
       "4        1  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.read_csv('full-gt.csv') # from URL: https://graphics.cs.msu.ru/projects/traffic-sign-recognition.html\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "1da9267c",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['autosave01_02_2012_09_13_32.jpg',\n",
       " 'autosave01_02_2012_09_13_33.jpg',\n",
       " 'autosave01_02_2012_09_13_34.jpg',\n",
       " 'autosave01_02_2012_09_13_35.jpg',\n",
       " 'autosave01_02_2012_09_13_36.jpg']"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "os.listdir('rtsd-frames')[:5] # from URL: https://www.kaggle.com/datasets/watchman/rtsd-dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dd0a8dfc",
   "metadata": {},
   "source": [
    "### Saving crop files"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "e2cf6b7e",
   "metadata": {},
   "outputs": [],
   "source": [
    "source_dir = 'rtsd-frames'\n",
    "target_dir = 'dataset'\n",
    "\n",
    "if not os.path.exists(target_dir):\n",
    "    os.makedirs(target_dir)\n",
    "\n",
    "def get_sign(\n",
    "        filename, x_from, y_from, width, height, sign_class, sign_id, \n",
    "        img_path=source_dir, res_path=target_dir\n",
    "    ): \n",
    "    img = Im.open(f'{img_path}/{filename}')\n",
    "    img = img.crop((x_from, y_from, x_from + width, y_from + height))\n",
    "    filename = f'{sign_class}___{sign_id}__{filename}'\n",
    "    img.save(f'{target_dir}/{filename}')\n",
    "    return {'filename': filename,  'sign_class': sign_class, 'sign_id': sign_id}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "7d81a1cf",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████████████████████████████████████████████████████████████████████| 104358/104358 [18:08<00:00, 95.90it/s]\n"
     ]
    }
   ],
   "source": [
    "result, bad_data = [], []\n",
    "for i in tqdm(range(len(data))):\n",
    "    try:\n",
    "        result.append(get_sign(**data.iloc[i].to_dict()))\n",
    "    except Exception as e:\n",
    "        bad_data.append((e, data.iloc[i].to_dict()))\n",
    "        print('.', end='')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "c288b477",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "104358"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(os.listdir(target_dir))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "51aae0a9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "104358"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "454cc4f4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(bad_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4fadacb7",
   "metadata": {},
   "source": [
    "### Metadata generation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "f9ee692c",
   "metadata": {},
   "outputs": [],
   "source": [
    "pd.DataFrame(result).to_csv(f'{target_dir}.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "07d9969e",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>file_name</th>\n",
       "      <th>additional_feature</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_33.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_34.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_35.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_36.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1_23___1__autosave01_02_2012_09_13_36.jpg</td>\n",
       "      <td>1_23</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                   file_name additional_feature\n",
       "0   2_1___0__autosave01_02_2012_09_13_33.jpg                2_1\n",
       "1   2_1___0__autosave01_02_2012_09_13_34.jpg                2_1\n",
       "2   2_1___0__autosave01_02_2012_09_13_35.jpg                2_1\n",
       "3   2_1___0__autosave01_02_2012_09_13_36.jpg                2_1\n",
       "4  1_23___1__autosave01_02_2012_09_13_36.jpg               1_23"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame(result)\n",
    "df.drop(columns=['sign_id'], inplace=True)\n",
    "df.columns = ['file_name', 'additional_feature']\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "57a8c62a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>file_name</th>\n",
       "      <th>additional_feature</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_33.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_34.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_35.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2_1___0__autosave01_02_2012_09_13_36.jpg</td>\n",
       "      <td>2_1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1_23___1__autosave01_02_2012_09_13_36.jpg</td>\n",
       "      <td>1_23</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                   file_name additional_feature\n",
       "0   2_1___0__autosave01_02_2012_09_13_33.jpg                2_1\n",
       "1   2_1___0__autosave01_02_2012_09_13_34.jpg                2_1\n",
       "2   2_1___0__autosave01_02_2012_09_13_35.jpg                2_1\n",
       "3   2_1___0__autosave01_02_2012_09_13_36.jpg                2_1\n",
       "4  1_23___1__autosave01_02_2012_09_13_36.jpg               1_23"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.to_json(f'{target_dir}/metadata.jsonl', orient='records', lines=True); df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "7245bea0",
   "metadata": {},
   "outputs": [],
   "source": [
    "metadata = pd.read_csv(f'{target_dir}.csv') # или metadata = df.copy()\n",
    "metadata.columns = ['image', 'sign_class', 'sign_id']\n",
    "metadata['image'] = metadata['image'].apply(lambda x: f'{target_dir}/{x}')\n",
    "metadata = metadata.to_dict(orient='list')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "591c55b5",
   "metadata": {},
   "source": [
    "### Creating a formatted dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "6d5fb041",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = Dataset.from_dict(metadata).cast_column(\"image\", Image())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "acd56ad2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Dataset({\n",
       "    features: ['image', 'sign_class', 'sign_id'],\n",
       "    num_rows: 104358\n",
       "})"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "98bb6893",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=27x26 at 0x1EFD1DD5C70>,\n",
       " 'sign_class': '2_1',\n",
       " 'sign_id': 0}"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset[2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "3ab455b9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=27x26 at 0x1EFD2DDBE80>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset[2]['image']"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "49420851",
   "metadata": {},
   "source": [
    "### Uploading to remote storage"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "1e3a67b4",
   "metadata": {},
   "outputs": [],
   "source": [
    "notebook_login()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "e59e80c5",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset.push_to_hub(\"eleldar/rtsd_cleaned\")"
   ]
  }
 ],
 "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
}