sharded-video-access.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import subprocess\n",
    "import tarfile\n",
    "import braceexpand"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "NB: Change these URLs to a copy of the dataset for training; please don't train against our storage bucket."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "url = \"http://storage.googleapis.com/lpr-yt8m-hi-sharded/yt8m-hi-000000.tar\"\n",
    "shards = \"http://storage.googleapis.com/lpr-yt8m-hi-sharded/yt8m-hi-{000000..000999}.tar\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def iterate_over_video_streams(url):\n",
    "    with subprocess.Popen([\"curl\", \"-s\", \"-L\", url], stdout=subprocess.PIPE) as proc:\n",
    "        with tarfile.TarFile.open(fileobj=proc.stdout, mode=\"r|\") as source:\n",
    "            for info in source:\n",
    "                if not info.name.endswith(\".mp4\"): continue\n",
    "                yield info, source.extractfile(info)\n",
    "                \n",
    "def write_stream(stream, fname):\n",
    "    with open(fname, \"wb\") as ostream:\n",
    "        while True:\n",
    "            data = stream.read(1000000)\n",
    "            if len(data) == 0: break\n",
    "            ostream.write(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "31521996 -3MxAiAnknY.mp4\n"
     ]
    }
   ],
   "source": [
    "for info, stream in iterate_over_video_streams(url):\n",
    "    print(info.size, info.name)\n",
    "    write_stream(stream, \"sample.mp4\")\n",
    "    # at this point, you can open and use \"sample.mp4\" just the way you usually process videos\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-rw-r--r-- 1 tmb tmb 31521996 Mar 23 18:09 sample.mp4\n"
     ]
    }
   ],
   "source": [
    "!ls -l sample.mp4\n",
    "# mplayer sample.mp4\n",
    "# etc."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To iterate over the entire dataset, use something like this:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for url in braceexpand.braceexpand(shards):\n",
    "    for info, stream in iterate_over_video_streams(url):\n",
    "        break\n",
    "    break"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}