Build_shape_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import random\n",
    "from os.path import isdir, join\n",
    "import menpo.io as mio"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Define the datasets to be loaded "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "p0 = '/vol/atlas/databases/'\n",
    "assert(isdir(p0))\n",
    "folds = ['ibug', 'afw', '300w', 'helen/trainset', \n",
    "         'helen/testset', 'lfpw/trainset', 'lfpw/testset']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load the actual landmark files "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "shapes = []\n",
    "for f in folds:    \n",
    "    sh1 = [ln for ln in mio.import_landmark_files(join(p0, f, ''))]\n",
    "    shapes = shapes + sh1\n",
    "    \n",
    "# below 'save' only the ones that have 68 points\n",
    "shapes68 = []\n",
    "for ln in shapes:\n",
    "    if ln.lms.points.shape[0] == 68:\n",
    "        # save only the ones with 68 points\n",
    "        shapes68.append(ln.lms)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# convert to 49 points from 68 mark-up.\n",
    "from menpo.landmark import face_ibug_68_to_face_ibug_49 as i68to49\n",
    "shapes68 = [i68to49(sh) for sh in shapes68]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Build ATM "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# shuffle the shapes\n",
    "random.Random(9).shuffle(shapes68)\n",
    "# build the ATM\n",
    "from menpofit.atm import HolisticATM\n",
    "# dummy image for template model\n",
    "im = mio.import_builtin_asset.einstein_jpg()\n",
    "if shapes68[0].n_points == 49:\n",
    "    # if we have transformed the landmarks to 49, the same should \n",
    "    # happen to the template landmarks.\n",
    "    im.landmarks['PTS'] = i68to49(im.landmarks['PTS'])\n",
    "atm = HolisticATM(im, shapes68, scales=[1.])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Visualise the ATM "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "try:\n",
    "    from menpowidgets.menpofit import visualize_atm\n",
    "    visualize_atm(atm)\n",
    "except:\n",
    "    # in case we call this from the terminal.\n",
    "    pass"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Export the ATM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# create a dictionary and export it.\n",
    "p_exp = '/vol/atlas/homes/grigoris/videos_external/christos_georgakis/2016_04/'\n",
    "assert(isdir(p_exp))\n",
    "from time import gmtime, strftime\n",
    "tm = strftime(\"%Y.%m.%d %H:%M:%S\", gmtime())\n",
    "clas1 = {'folds' : folds, 'date' : tm, 'path_base' : p0, 'atm': atm}\n",
    "mio.export_pickle(clas1, p_exp + 'shape_model_tt_49.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Export in mat format (only PCA) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "c = atm.shape_models[0].model\n",
    "components = c.components\n",
    "eigenvalues = c.eigenvalues\n",
    "eig_cr = c.eigenvalues_cumulative_ratio()\n",
    "\n",
    "from scipy.io import savemat\n",
    "savemat(p_exp + 'shape_model_tt_49.mat', {'components': components, 'eigenvalues' : eigenvalues, \n",
    "                                   'eigenvalues_cumulative_ratio' : eig_cr})"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}