Bad Frame Plotting.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import boto3\n",
    "import sagemaker\n",
    "import s3fs\n",
    "fs = s3fs.S3FileSystem()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "BUCKET = 'pose-estimates-for-bodhi' # the name of the bucket containing the pose jsons"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Example code - finding bad frames.ipynb\n",
      "Mouse#20_BalbC_20180423_13-28-05_Front.seq\n",
      "Mouse#20_BalbC_20180423_13-28-05_Top.seq\n",
      "Mouse#20_BalbC_20180423_13-28-05_pose_top_v1_5.json\n",
      "Mouse#20_CastBalbC_20180423_11-37-37_Front.seq\n",
      "Mouse#20_CastBalbC_20180423_11-37-37_Top.seq\n",
      "Mouse#20_CastBalbC_20180423_11-37-37_pose_top_v1_5.json\n",
      "Mouse#22_BalbC_20180423_12-11-55_Front.seq\n",
      "Mouse#22_BalbC_20180423_12-11-55_Top.seq\n",
      "Mouse#22_BalbC_20180423_12-11-55_pose_top_v1_5.json\n",
      "Mouse#23_BalbC_20180423_12-39-09_Front.seq\n",
      "Mouse#23_BalbC_20180423_12-39-09_Top.seq\n",
      "Mouse#23_BalbC_20180423_12-39-09_pose_top_v1_5.json\n",
      "Mouse#24_BalbC_20180423_13-47-49_Front.seq\n",
      "Mouse#24_BalbC_20180423_13-47-49_Top.seq\n",
      "Mouse#24_BalbC_20180423_13-47-49_pose_top_v1_5.json\n",
      "Mouse#25_BalbC_20180423_14-30-32_Front.seq\n",
      "Mouse#25_BalbC_20180423_14-30-32_Top.seq\n",
      "Mouse#25_BalbC_20180423_14-30-32_pose_top_v1_5.json\n",
      "Mouse1100_20180110_20-51-38_Front.seq\n",
      "Mouse1100_20180110_20-51-38_Top.seq\n",
      "Mouse1100_20180110_20-51-38_pose_top_v1_5.json\n",
      "Mouse1100_20180110_21-59-59_Front.seq\n",
      "Mouse1100_20180110_21-59-59_Top.seq\n",
      "Mouse1100_20180110_21-59-59_pose_top_v1_5.json\n",
      "Mouse1102_20180117_20-22-25_Front.seq\n",
      "Mouse1102_20180117_20-22-25_Top.seq\n",
      "Mouse1102_20180117_20-22-25_pose_top_v1_5.json\n",
      "Mouse1139_20180406_11-19-25_Front.seq\n",
      "Mouse1139_20180406_11-19-25_Top.seq\n",
      "Mouse1139_20180406_11-19-25_pose_top_v1_5.json\n",
      "Mouse1143_20180405_13-08-57_Front.seq\n",
      "Mouse1143_20180405_13-08-57_Top.seq\n",
      "Mouse1143_20180405_13-08-57_pose_top_v1_5.json\n",
      "Mouse1145_20180405_12-51-10_Front.seq\n",
      "Mouse1145_20180405_12-51-10_Top.seq\n",
      "Mouse1145_20180405_12-51-10_pose_top_v1_5.json\n",
      "Mouse287_20170713_18-22-43_Front.seq\n",
      "Mouse287_20170713_18-22-43_Top.seq\n",
      "Mouse287_20170713_18-22-43_pose_top_v1_5.json\n",
      "Mouse291_20170711_15-27-55_Front.seq\n",
      "Mouse291_20170711_15-27-55_Top.seq\n",
      "Mouse291_20170711_15-27-55_pose_top_v1_5.json\n",
      "Mouse351_20171115_16-05-01_Front_J85.seq\n",
      "Mouse351_20171115_16-05-01_Top_J85.seq\n",
      "Mouse351_20171115_16-05-01_pose_top_v1_5.json\n",
      "Mouse447_20180117_12-35-57_Front_J85.seq\n",
      "Mouse447_20180117_12-35-57_Top_J85.seq\n",
      "Mouse447_20180117_12-35-57_pose_top_v1_5.json\n",
      "Mouse479_20180215_14-15-58_Front_J85.seq\n",
      "Mouse479_20180215_14-15-58_Top_J85.seq\n",
      "Mouse479_20180215_14-15-58_pose_top_v1_5.json\n",
      "Mouse482_20180314_13-39-24_Front_J85.seq\n",
      "Mouse482_20180314_13-39-24_Top_J85.seq\n",
      "Mouse482_20180314_13-39-24_pose_top_v1_5.json\n",
      "Mouse513_female_20180404_16-09-17_Front_J85.seq\n",
      "Mouse513_female_20180404_16-09-17_Top_J85.seq\n",
      "Mouse513_female_20180404_16-09-17_pose_top_v1_5.json\n",
      "Mouse513_male_20180404_15-42-09_Front_J85.seq\n",
      "Mouse513_male_20180404_15-42-09_Top_J85.seq\n",
      "Mouse513_male_20180404_15-42-09_pose_top_v1_5.json\n",
      "Mouse515_male_20180404_12-38-38_Front_J85.seq\n",
      "Mouse515_male_20180404_12-38-38_Top_J85.seq\n",
      "Mouse515_male_20180404_12-38-38_pose_top_v1_5.json\n",
      "Mouse524_20180521_17-12-38_Front.seq\n",
      "Mouse524_20180521_17-12-38_Top.seq\n",
      "Mouse524_20180521_17-12-38_pose_top_v1_7.json\n",
      "Mouse524_20180521_17-26-50_Front.seq\n",
      "Mouse524_20180521_17-26-50_Top.seq\n",
      "Mouse524_20180521_17-26-50_pose_top_v1_7.json\n",
      "Mouse524_20180521_17-46-31_Front.seq\n",
      "Mouse524_20180521_17-46-31_Top.seq\n",
      "Mouse524_20180521_17-46-31_pose_top_v1_7.json\n",
      "Mouse524_20180521_18-01-31_Front.seq\n",
      "Mouse524_20180521_18-01-31_Top.seq\n",
      "Mouse524_20180521_18-01-31_pose_top_v1_7.json\n",
      "Mouse529_20180521_18-40-59_Front.seq\n",
      "Mouse529_20180521_18-40-59_Top.seq\n",
      "Mouse529_20180521_18-40-59_pose_top_v1_7.json\n",
      "Mouse529_20180521_19-11-21_Front.seq\n",
      "Mouse529_20180521_19-11-21_Top.seq\n",
      "Mouse529_20180521_19-11-21_pose_top_v1_7.json\n",
      "Mouse529_20180521_19-38-42_Front.seq\n",
      "Mouse529_20180521_19-38-42_Top.seq\n",
      "Mouse529_20180521_19-38-42_pose_top_v1_7.json\n",
      "Mouse_20170302_10-50-34_Front_J85.seq\n",
      "Mouse_20170302_10-50-34_Top_J85.seq\n",
      "Mouse_20170302_10-50-34_pose_top_v1_5.json\n",
      "Mouse_20170302_12-39-09_Front_J85.seq\n",
      "Mouse_20170302_12-39-09_Top_J85.seq\n",
      "Mouse_20170302_12-39-09_pose_top_v1_5.json\n",
      "Mouse_20170314_14-19-03_Front_J85.seq\n",
      "Mouse_20170314_14-19-03_Top_J85.seq\n",
      "Mouse_20170314_14-19-03_pose_top_v1_5.json\n",
      "Mouse_20170321_10-59-36_Front_J85.seq\n",
      "Mouse_20170321_10-59-36_Top_J85.seq\n",
      "Mouse_20170321_10-59-36_pose_top_v1_5.json\n",
      "Mouse_20170324_12-12-09_Front_J85.seq\n",
      "Mouse_20170324_12-12-09_Top_J85.seq\n",
      "Mouse_20170324_12-12-09_pose_top_v1_5.json\n",
      "Mouse_20170329_08-37-39_Front_J85.seq\n",
      "Mouse_20170329_08-37-39_Top_J85.seq\n",
      "Mouse_20170329_08-37-39_pose_top_v1_5.json\n",
      "Mouse_20170719_13-08-14_Front.seq\n",
      "Mouse_20170719_13-08-14_Top.seq\n",
      "Mouse_20170719_13-08-14_pose_top_v1_5.json\n",
      "Mouse_20170906_11-08-18_Front_J85.seq\n",
      "Mouse_20170906_11-08-18_Top_J85.seq\n",
      "Mouse_20170906_11-08-18_pose_top_v1_5.json\n",
      "Mouse_20170906_16-02-55_Front_J85.seq\n",
      "Mouse_20170906_16-02-55_Top_J85.seq\n",
      "Mouse_20170906_16-02-55_pose_top_v1_5.json\n",
      "Mouse_20170928_11-12-53_Front_J85.seq\n",
      "Mouse_20170928_11-12-53_Top_J85.seq\n",
      "Mouse_20170928_11-12-53_pose_top_v1_5.json\n",
      "Mouse_20170928_13-06-53_Front_J85.seq\n",
      "Mouse_20170928_13-06-53_Top_J85.seq\n",
      "Mouse_20170928_13-06-53_pose_top_v1_5.json\n",
      "Mouse_20180116_15-02-15_Front_J85.seq\n",
      "Mouse_20180116_15-02-15_Top_J85.seq\n",
      "Mouse_20180116_15-02-15_pose_top_v1_5.json\n",
      "Mouse_20180116_15-34-50_Front_J85.seq\n",
      "Mouse_20180116_15-34-50_Top_J85.seq\n",
      "Mouse_20180116_15-34-50_pose_top_v1_5.json\n",
      "Mouse_20180222_18-06-26_Front.seq\n",
      "Mouse_20180222_18-06-26_Top.seq\n",
      "Mouse_20180222_18-06-26_pose_top_v1_5.json\n",
      "Mouse_20180302_18-50-52_Front.seq\n",
      "Mouse_20180302_18-50-52_Top.seq\n",
      "Mouse_20180302_18-50-52_pose_top_v1_5.json\n",
      "saved_frames/\n",
      "saved_frames/Mouse#20_BalbC_20180423_13-28-05_Top_1234.jpg\n",
      "seqIo.py\n"
     ]
    }
   ],
   "source": [
    "for key in s3.list_objects(Bucket = BUCKET)['Contents']:\n",
    "    print(key['Key'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "role = sagemaker.get_execution_role()\n",
    "region = boto3.session.Session().region_name\n",
    "s3 = boto3.client('s3')\n",
    "\n",
    "bucket_region = s3.head_bucket(Bucket=BUCKET)['ResponseMetadata']['HTTPHeaders']['x-amz-bucket-region']\n",
    "assert bucket_region == region, \"You S3 bucket {} and this notebook need to be in the same region.\".format(BUCKET)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# example code- how to load a .json from an s3 bucket into Python:\n",
    "\n",
    "# get all jsons in the bucket.\n",
    "json_list = fs.glob('s3://{}/*.json'.format(BUCKET))\n",
    "\n",
    "# take the first one as an example.\n",
    "fid = 's3://' + json_list[35]\n",
    "\n",
    "with fs.open(fid,'rb') as file:\n",
    "    data = json.load(file)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The dimensions of `data['keypoints']` are (frames) x (mouse number) x (x,y) x (body part).\n",
    "\n",
    "Body parts are ordered (nose, ear1, ear2, neck, hip1, hip2, tail).\n",
    "\n",
    "some examples:\n",
    "- `data['keypoints'][100][0]` gives all keypoints from the first mouse on the 101st (counting from 0) frame.\n",
    "- `data['keypoints'][100][1][0][2]` gives the x-coordinate of ear2 from the second mouse on the 101st frame.\n",
    "\n",
    "A few other notes:\n",
    "- Values are in pixels, with a typical mouse being (I think) ~350 pixels long.\n",
    "- Sometimes the white mouse is not present at the start of recording: you will be able to tell this because all 7 keypoints will be very small values a few pixels apart.\n",
    "- In videos that start out with the white mouse being absent, the mouse is lowered into the cage by an experimenter. While the experimenter's hand is present, both mice are often obscured and the pose estimates will jump all over the place- but these frames shouldn't be pulled out as error frames, since human annotators won't know how to label them either!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# if we plot the body parts in this order we'll get a figure-8 mouse pose:\n",
    "point_order = np.array([3, 2, 0, 1, 3, 4, 6, 5, 3])\n",
    "\n",
    "\n",
    "f1 = 16280\n",
    "\n",
    "# load the x,y coordinates of all body parts of the first mouse on frame f1 and frame f1+1.\n",
    "# We convert the data into a numpy array format to make it easier to work with.\n",
    "frame1 = np.array(data['keypoints'][f1][0])\n",
    "frame2 = np.array(data['keypoints'][f1+1][0])\n",
    "\n",
    "\n",
    "# plot where the first mouse is on frame 1, with an x marking the nose.\n",
    "plt.plot(frame1[0][point_order], frame1[1][point_order],'.-')\n",
    "plt.plot(frame1[0][0], frame1[1][0],'x')\n",
    "\n",
    "# plot where the first mouse is on frame 2, with an x marking the nose.\n",
    "plt.plot(frame2[0][point_order], frame2[1][point_order],'.-')\n",
    "plt.plot(frame2[0][0], frame2[1][0],'x')\n",
    "\n",
    "# add a legend, set axis limits, and display the plot.\n",
    "plt.legend(('frame 1 keypoint locations','frame 1 nose','frame 2 keypoint locations','frame 2 nose'))\n",
    "plt.xlim(0, 1300)\n",
    "plt.ylim(0, 600)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "837.4402627661171\n"
     ]
    }
   ],
   "source": [
    "# as you can see, the pose estimates on these two frames are pretty close together,\n",
    "# suggesting that pose estimation is working fine.\n",
    "\n",
    "# To measure how much the mouse moved between these two frames, we will compute the\n",
    "# mean Euclidean Distance between each keypoint on frame 1 and each keypoint on frame 2:\n",
    "\n",
    "dX = frame2[0] - frame1[0]\n",
    "dY = frame2[1] - frame1[1]\n",
    "total_change = np.mean(np.sqrt(dX**2 + dY**2)) # **2 squares all the entries of dX.\n",
    "\n",
    "# displays the total change in keypoint position, in pixels, between these two frames:\n",
    "print(total_change)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "conda_python3",
   "language": "python",
   "name": "conda_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": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}