{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. PP-TinyPose模型简介\n",
    "PP-TinyPose是PaddleDetecion针对移动端设备优化的实时关键点检测模型，可流畅地在移动端设备上执行多人姿态估计任务。借助PaddleDetecion自研的优秀轻量级检测模型[PicoDet](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/configs/picodet/README.md)，我们同时提供了特色的轻量级垂类行人检测模型.TinyPose的运行环境有以下依赖要求：\n",
    "\n",
    "PaddlePaddle>=2.2\n",
    "\n",
    "如希望在移动端部署，则还需要：\n",
    "\n",
    "Paddle-Lite>=2.11"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "更多部署案例可参考[PP-TinyPose](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/configs/keypoint/tiny_pose/README.md)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. 模型效果及应用场景\n",
    "### 2.1 关键点检测任务：\n",
    "\n",
    "#### 2.1.1 数据集：\n",
    "\n",
    "目前KeyPoint模型支持[COCO](https://cocodataset.org/#keypoints-2017)数据集和[MPII](http://human-pose.mpi-inf.mpg.de/#overview)数据集，数据集的准备方式请参考[关键点数据准备](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/docs/tutorials/data/PrepareKeypointDataSet.md)\n",
    "\n",
    "#### 2.1.2 模型效果速览：\n",
    "\n",
    "PP-TinyPose的检测效果为：\n",
    "\n",
    "![](https://user-images.githubusercontent.com/15810355/181733705-d0f84232-c6a2-43dd-be70-4a3a246b8fbc.gif)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. 模型如何使用\n",
    "\n",
    "### 3.1 模型推理：\n",
    "* 下载 \n",
    "\n",
    "（不在Jupyter Notebook上运行时需要将\"！\"或者\"%\"去掉。）\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "jupyter": {
     "outputs_hidden": false
    },
    "scrolled": true,
    "tags": []
   },
   "outputs": [],
   "source": [
    "# 克隆PaddleDetection仓库\n",
    "%mkdir -p ~/work\n",
    "%cd ~/work/\n",
    "!git clone https://github.com/PaddlePaddle/PaddleDetection.git\n",
    "\n",
    "# 安装其他依赖\n",
    "%cd PaddleDetection\n",
    "%mkdir -p demo_input demo_output\n",
    "!pip install -r requirements.txt\n",
    "\n",
    "# 开始安装PaddleDetection \n",
    "!python setup.py install  #如果安装过程中长时间卡住，可中断后继续重新执行"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* 验证是否安装成功\n",
    "如果报错，只需执行上一步操作。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# 测试是否安装成功\n",
    "!python ppdet/modeling/tests/test_architectures.py"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* 快速体验\n",
    "\n",
    "恭喜！ 您已经成功安装了PaddleDetection，接下来快速关键点检测效果。您可以直接下载模型库中提供的对应预测部署模型，分别获取得到行人检测模型和关键点检测模型的预测部署模型，解压即可。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true,
    "tags": []
   },
   "outputs": [],
   "source": [
    "# 下载模型\n",
    "!mkdir -p output_inference\n",
    "%cd output_inference\n",
    "# 下载行人检测模型s\n",
    "!wget https://bj.bcebos.com/v1/paddledet/models/keypoint/tinypose_enhance/picodet_s_320_lcnet_pedestrian.zip\n",
    "!unzip picodet_s_320_lcnet_pedestrian.zip\n",
    "# 下载关键点检测模型\n",
    "!wget https://bj.bcebos.com/v1/paddledet/models/keypoint/tinypose_enhance/tinypose_128x96.zip\n",
    "!unzip tinypose_128x96.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%cd ~/work/PaddleDetection/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 预测一张图片\n",
    "!wget -P demo_input -N https://paddledet.bj.bcebos.com/modelcenter/images/PP-TinyPose/000000568213.jpg\n",
    "!python deploy/python/det_keypoint_unite_infer.py --det_model_dir=output_inference/picodet_v2_s_320_pedestrian --keypoint_model_dir=output_inference/tinypose_128x96 --image_file=demo_input/000000568213.jpg --device=GPU --output_dir=demo_output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 预测一个视频\n",
    "!wget -P demo_input -N https://paddledet.bj.bcebos.com/modelcenter/images/PP-TinyPose/demo_PP-TinyPose.mp4\n",
    "!python deploy/python/det_keypoint_unite_infer.py --det_model_dir=output_inference/picodet_v2_s_320_pedestrian --keypoint_model_dir=output_inference/tinypose_128x96 --video_file=demo_input/demo_PP-TinyPose.mp4 --device=GPU --output_dir=demo_output"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.2 模型训练：\n",
    "* 克隆PaddleDetection仓库（详见3.1）"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* 数据集准备\n",
    "\n",
    "    关键点检测模型与行人检测模型的训练集在`COCO`以外还扩充了[AI Challenger](https://arxiv.org/abs/1711.06475)数据集，各数据集关键点定义如下：\n",
    "    ```\n",
    "    COCO keypoint Description:\n",
    "        0: \"Nose\",\n",
    "        1: \"Left Eye\",\n",
    "        2: \"Right Eye\",\n",
    "        3: \"Left Ear\",\n",
    "        4: \"Right Ear\",\n",
    "        5: \"Left Shoulder,\n",
    "        6: \"Right Shoulder\",\n",
    "        7: \"Left Elbow\",\n",
    "        8: \"Right Elbow\",\n",
    "        9: \"Left Wrist\",\n",
    "        10: \"Right Wrist\",\n",
    "        11: \"Left Hip\",\n",
    "        12: \"Right Hip\",\n",
    "        13: \"Left Knee\",\n",
    "        14: \"Right Knee\",\n",
    "        15: \"Left Ankle\",\n",
    "        16: \"Right Ankle\"\n",
    "\n",
    "    AI Challenger Description:\n",
    "        0: \"Right Shoulder\",\n",
    "        1: \"Right Elbow\",\n",
    "        2: \"Right Wrist\",\n",
    "        3: \"Left Shoulder\",\n",
    "        4: \"Left Elbow\",\n",
    "        5: \"Left Wrist\",\n",
    "        6: \"Right Hip\",\n",
    "        7: \"Right Knee\",\n",
    "        8: \"Right Ankle\",\n",
    "        9: \"Left Hip\",\n",
    "        10: \"Left Knee\",\n",
    "        11: \"Left Ankle\",\n",
    "        12: \"Head top\",\n",
    "        13: \"Neck\"\n",
    "    ```\n",
    "\n",
    "    由于两个数据集的关键点标注形式不同，我们将两个数据集的标注进行了对齐，仍然沿用COCO的标注形式，您可以下载[训练的参考列表](https://bj.bcebos.com/v1/paddledet/data/keypoint/aic_coco_train_cocoformat.json)并放在`dataset/`下使用。对齐两个数据集标注文件的主要处理如下：\n",
    "    - `AI Challenger`关键点标注顺序调整至与COCO一致，统一是否标注/可见的标志位；\n",
    "    - 舍弃了`AI Challenger`中特有的点位；将`AI Challenger`数据中`COCO`特有点位标记为未标注；\n",
    "    - 重新排列了`image_id`与`annotation id`；\n",
    "    利用转换为`COCO`形式的合并数据标注，执行模型训练。\n",
    "    若用户需要自定义数据集，可参考[快速开始-自定义数据集](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/keypoint#%E5%BF%AB%E9%80%9F%E5%BC%80%E5%A7%8B)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 关键点检测模型\n",
    "!python -m paddle.distributed.launch tools/train.py -c configs/keypoint/tiny_pose/tinypose_128x96.yml"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 行人检测模型\n",
    "!python -m paddle.distributed.launch tools/train.py -c configs/picodet/application/pedestrian_detection/picodet_s_320_lcnet_pedestrian.yml"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. 方案介绍\n",
    "<div align=\"center\">\n",
    "  <img src=\"tinypose_pipeline.png\" width='800'/>\n",
    "</div>"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.8.13 ('paddle_env')",
   "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": "3.8.13"
  },
  "vscode": {
   "interpreter": {
    "hash": "864bc28e4d94d9c1c4bd0747e4313c0ab41718ab445ced17dbe1a405af5ecc64"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}