{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "source": [
    "import jsonlines\n",
    "import os"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "source": [
    "import cv2"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "source": [
    "# extract first frame from mp4 file and save it as image\n",
    "\n",
    "def extract_frame(mp4,file_name):\n",
    "    save_path='./imgs/'+file_name[:-4]+\".jpg\"\n",
    "    # print(save_path)\n",
    "    if os.path.exists(save_path):\n",
    "        # print(\"file already exists\")\n",
    "        return\n",
    "    else:\n",
    "        try:\n",
    "            # extract first frame\n",
    "            cap = cv2.VideoCapture(mp4)\n",
    "            ret, frame = cap.read()\n",
    "            cv2.imwrite(save_path, frame)\n",
    "        except:\n",
    "            print(f\"error in {save_path} file\")"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "source": [
    "# read jsonlines\n",
    "im_un=[]\n",
    "base_dir=\"./gifs\"\n",
    "data=jsonlines.open('./ReactionGIF.json')\n",
    "\n",
    "train_writer=jsonlines.open(\"train.json\", mode='w')\n",
    "val_writer=jsonlines.open(\"val.json\", mode='w')\n",
    "writer=train_writer\n",
    "\n",
    "for x in data:\n",
    "    text=x[\"text\"]\n",
    "    sentiment=x[\"label\"]\n",
    "    image_name=x[\"reply\"]\n",
    "    if image_name is None:\n",
    "        continue\n",
    "    else:\n",
    "        image_name=image_name.split(\"/\")[-1]\n",
    "    jpg_name='/home/ceyda/data/ReactionGIF/imgs/'+image_name[:-4]+\".jpg\"\n",
    "\n",
    "    # image_path=os.path.join(base_dir,image_name)\n",
    "    # extract_frame(image_path,image_name)\n",
    "    if os.path.exists(jpg_name):\n",
    "        dic={\n",
    "            \"image_path\":jpg_name,\n",
    "            \"captions\":[text,sentiment]\n",
    "        }\n",
    "        im_un.append(jpg_name)\n",
    "        if len(im_un)>=19000:\n",
    "            writer=val_writer\n",
    "        writer.write(dic)\n",
    "        # text,jpg_name,sentiment"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "source": [
    "len(im_un)"
   ],
   "outputs": [
    {
     "output_type": "execute_result",
     "data": {
      "text/plain": [
       "19387"
      ]
     },
     "metadata": {},
     "execution_count": 13
    }
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "source": [
    "len(set(im_un))"
   ],
   "outputs": [
    {
     "output_type": "execute_result",
     "data": {
      "text/plain": [
       "19385"
      ]
     },
     "metadata": {},
     "execution_count": 14
    }
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "source": [
    "import jax"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "source": [
    "jax.device_count()"
   ],
   "outputs": [
    {
     "output_type": "execute_result",
     "data": {
      "text/plain": [
       "8"
      ]
     },
     "metadata": {},
     "execution_count": 18
    }
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "source": [
    "import shutil\n",
    "from pathlib import Path\n",
    "val=jsonlines.open('./val.json')\n",
    "for v in val:\n",
    "\n",
    "    shutil.copy(v['image_path'],\"/home/ceyda/code/clip-reply-demo/imgs/\" +Path(v['image_path']).name)\n"
   ],
   "outputs": [],
   "metadata": {}
  }
 ],
 "metadata": {
  "orig_nbformat": 4,
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}