{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Download YOLOv3 and SimplePose model from MXNET model zoo\n",
    "- Choose Data Science kernel and ml.t3.medium instance\n",
    "- Run following code to have models download\n",
    "- The downloaded models are saved in ~/.mxnet/models folder. You should be able to see them from the Image Terminal\n",
    "- Go into the ~/.mxnet/models folder and use following command to create a tar ball file including these two models  \n",
    "`\n",
    "cd ~/.mxnet/models\n",
    "tar -cvzf model.tar.gz simple_pose_resnet18_v1b-f63d42ac.params yolo3_mobilenet1.0_coco-66dbbae6.params\n",
    "`\n",
    "- Upload the model.tar.gz file into an S3 bucket with following command. Modify S3_BUCKET_NAME and MODEL_FILE_KEY_PREFIX to your own choice.  \n",
    "`aws s3 cp model.tar.gz s3://{S3_BUCKET_NAME}/{MODEL_FILE_KEY_PREFIX}/model.tar.gz`\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gluoncv import model_zoo, data, utils\n",
    "detector = model_zoo.get_model('yolo3_mobilenet1.0_coco', pretrained=True)\n",
    "pose_net = model_zoo.get_model('simple_pose_resnet18_v1b', pretrained=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Build with original MXNet Docker Image\n",
    "- Run following code to create a SageMaker model with the model file\n",
    "- A SageMaker endpoint will be created based on the MXNet model file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import json\n",
    "import boto3\n",
    "from time import gmtime, strftime\n",
    "client = boto3.client('sagemaker-runtime')\n",
    "import sagemaker\n",
    "from sagemaker.mxnet import MXNetModel\n",
    "bucket_name = 'S3_BUCKET_NAME'\n",
    "s3_prefix = 'MODEL_FILE_KEY_PREFIX'\n",
    "model = MXNetModel(\n",
    "    entry_point='entrypoint.py',\n",
    "    model_data='s3://{}/{}/model.tar.gz'.format(bucket_name, s3_prefix),\n",
    "    framework_version='1.6.0',\n",
    "    py_version='py3',\n",
    "    source_dir='inference_src',\n",
    "    role=sagemaker.get_execution_role(),\n",
    "    name='Extract-pose-keypoints-yolo3-simplepose-'+ strftime('%Y-%m-%d-%H-%M-%S', gmtime())\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "ep_name = 'Extract-pose-keypoints-yolo3-simplepose-'+ strftime('%Y-%m-%d-%H-%M-%S', gmtime())\n",
    "predictor = model.deploy(initial_instance_count=1, \n",
    "                        instance_type='ml.m5.4xlarge',\n",
    "                        endpoint_name=ep_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inference sample code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Test with SageMaker runtime client\n",
    "test_image = 'images/soccer.png'\n",
    "img = open(test_image, 'rb').read()\n",
    "response = client.invoke_endpoint(\n",
    "    EndpointName=ep_name,\n",
    "    Body=img,\n",
    "    ContentType='application/x-image')\n",
    "\n",
    "result = response[\"Body\"].read().decode(\"utf-8\")\n",
    "detections = json.loads(result)\n",
    "print (detections)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Test with SageMaker predictor\n",
    "from sagemaker.predictor import Predictor\n",
    "\n",
    "object_detector = Predictor(endpoint_name=ep_name)\n",
    "results = object_detector.predict(img,initial_args={\"ContentType\": \"image/jpeg\"})\n",
    "detections = json.loads(results)\n",
    "detections"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Show keypoints"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "img = plt.imread(test_image)\n",
    "print(img.shape)\n",
    "fig,ax = plt.subplots(1, dpi=120)\n",
    "ax.imshow(img)\n",
    "for dots in detections['1']['qcresult']:\n",
    "    for idx, dot in enumerate(dots):\n",
    "        if idx==0: continue\n",
    "        x,y,xb,yb = dot\n",
    "        x = x*img.shape[1]\n",
    "        y = y*img.shape[0]\n",
    "           \n",
    "        rect = plt.plot(x,y,'b.')\n",
    "ax.axis('off')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Delete the endpoint after use"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "boto3.client('sagemaker').delete_endpoint(EndpointName=ep_name)"
   ]
  }
],
 "metadata": {
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}