{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "tags": [
     "remove-cell"
    ]
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "import openpifpaf\n",
    "openpifpaf.show.Canvas.show = True"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Datasets\n",
    "\n",
    "This section focuses on the COCO keypoint dataset which was the original\n",
    "dataset that OpenPifPaf started with. In general, \n",
    "training datasets are large and require a computer with a good GPU to train and \n",
    "evaluate in reasonable times. Additional datasets are availble as plugins \n",
    "(for example {doc}`plugins_crowdpose`, {doc}`plugins_wholebody`, \n",
    "{doc}`plugins_apollocar3d` and {doc}`plugins_animalpose`).\n",
    "\n",
    "```{note}\n",
    "These datasets are not required to do pose predictions on your own images.\n",
    "Even for training, you are unlikely to need all the datasets for your use case.\n",
    "```\n",
    "\n",
    "OpenPifPaf is extendible with plugins and it has been our focus to make \n",
    "it particularly easy to extend it with custom datasets that are formatted\n",
    "in the COCO format. Please see the tutorial on {doc}`custom datasets <plugins_custom>`\n",
    "for a step-by-step walkthrough."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Download COCO data\n",
    "\n",
    "[COCO](http://cocodataset.org/) is a great datasets containing many types of annotations, including bounding boxes, 2D poses, etc.\n",
    "You can also copy this code block into a code cell in Jupyter or Google Colab with the prefix `%%bash`.\n",
    "\n",
    "```sh\n",
    "mkdir data-mscoco\n",
    "cd data-mscoco\n",
    "\n",
    "wget -q -nc http://images.cocodataset.org/annotations/annotations_trainval2017.zip\n",
    "wget -q -nc http://images.cocodataset.org/annotations/image_info_test2017.zip\n",
    "unzip -q -n annotations_trainval2017.zip\n",
    "unzip -q -n image_info_test2017.zip\n",
    "\n",
    "mkdir images\n",
    "cd images\n",
    "wget -q -nc http://images.cocodataset.org/zips/val2017.zip\n",
    "wget -q -nc http://images.cocodataset.org/zips/train2017.zip\n",
    "wget -q -nc http://images.cocodataset.org/zips/test2017.zip\n",
    "unzip -q -n val2017.zip\n",
    "unzip -q -n train2017.zip\n",
    "unzip -q -n test2017.zip\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## COCO Person Skeletons\n",
    "\n",
    "COCO / kinematic tree / dense:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "tags": [
     "hide-input"
    ]
   },
   "outputs": [],
   "source": [
    "# HIDE CODE\n",
    "\n",
    "# first make an annotation\n",
    "ann_coco = openpifpaf.Annotation.from_cif_meta(\n",
    "    openpifpaf.plugins.coco.CocoKp().head_metas[0])\n",
    "ann_kin = openpifpaf.Annotation.from_cif_meta(\n",
    "    openpifpaf.plugins.coco.CocoKp(skeleton=openpifpaf.plugins.coco.constants.KINEMATIC_TREE_SKELETON).head_metas[0])\n",
    "ann_dense = openpifpaf.Annotation.from_cif_meta(\n",
    "    openpifpaf.plugins.coco.CocoKp(skeleton=openpifpaf.plugins.coco.constants.DENSER_COCO_PERSON_SKELETON).head_metas[0])\n",
    "\n",
    "# visualize the annotation\n",
    "openpifpaf.show.KeypointPainter.show_joint_scales = True\n",
    "keypoint_painter = openpifpaf.show.KeypointPainter()\n",
    "with openpifpaf.show.Canvas.annotation(ann_coco, ncols=3) as (ax1, ax2, ax3):\n",
    "    keypoint_painter.annotation(ax1, ann_coco)\n",
    "    keypoint_painter.annotation(ax2, ann_kin)\n",
    "    keypoint_painter.annotation(ax3, ann_dense)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "(coco-person-keypoints)=\n",
    "## COCO Person Keypoints"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i, name in enumerate(openpifpaf.plugins.coco.constants.COCO_KEYPOINTS):\n",
    "    print(i, name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print('associations')\n",
    "kp_names = openpifpaf.plugins.coco.constants.COCO_KEYPOINTS\n",
    "for i, (joint1, joint2) in enumerate(openpifpaf.plugins.coco.constants.COCO_PERSON_SKELETON):\n",
    "    print('{:2d}: {:15s} --> {}'.format(i, kp_names[joint1 - 1], kp_names[joint2 - 1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Download MPII data [draft]\n",
    "\n",
    "This MPII data is currently not used anywhere.\n",
    "\n",
    "```sh\n",
    "mkdir data-mpii\n",
    "cd data-mpii\n",
    "wget https://datasets.d2.mpi-inf.mpg.de/andriluka14cvpr/mpii_human_pose_v1.tar.gz\n",
    "wget https://datasets.d2.mpi-inf.mpg.de/andriluka14cvpr/mpii_human_pose_v1_u12_2.zip\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Download NightOwls data [draft]\n",
    "\n",
    "```sh\n",
    "mkdir data-nightowls\n",
    "cd data-nightowls\n",
    "wget http://www.robots.ox.ac.uk/\\~vgg/data/nightowls/python/nightowls_validation.json\n",
    "wget http://www.robots.ox.ac.uk/\\~vgg/data/nightowls/python/nightowls_validation.zip\n",
    "unzip nightowls_validation.zip\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "ea6946363a43e80d241452ab397f4c58bdd3d2517da174158e9c46ce6717422a"
  },
  "kernelspec": {
   "display_name": "Python 3.9.4 64-bit ('venv3': venv)",
   "name": "python3"
  },
  "language_info": {
   "name": "python",
   "version": ""
  },
  "metadata": {
   "interpreter": {
    "hash": "ea6946363a43e80d241452ab397f4c58bdd3d2517da174158e9c46ce6717422a"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}