Co-occurrence Feature Learning for Action Recognition.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/fan/anaconda3/envs/keras/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import math\n",
    "from keras import optimizers\n",
    "from utils import *\n",
    "from model import *"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Download the raw dataset if necessary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "SBU_dir ='/home/fan/projects/skeleton/data/SBU/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Download dataset\n",
    "'''\n",
    "link = []\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s01s02.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s01s03.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s01s07.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s02s01.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s02s03.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s02s06.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s02s07.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s03s02.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s03s04.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s03s05.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s03s06.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s04s02.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s04s03.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s04s06.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s05s02.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s05s03.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s06s02.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s06s03.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s06s04.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s07s01.zip')\n",
    "link.append('http://vision.cs.stonybrook.edu/~kiwon/Datasets/SBU_Kinect_Interactions/s07s03.zip')\n",
    "\n",
    "import wget\n",
    "for i in link:\n",
    "    temp_path = SBU_dir + i.split('/')[-1]\n",
    "    url = i\n",
    "    print(url)\n",
    "    print(temp_path)\n",
    "    wget.download(url=url, out=temp_path)\n",
    "'''"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Assign the dataset by choose folder 3 as test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "loading data from: /home/fan/projects/skeleton/data/SBU/\n",
      "test set folder should be slected from 0 ~ 4\n",
      "slected test folder 3 includes: ['s02s01', 's02s06', 's03s02', 's03s06']\n"
     ]
    }
   ],
   "source": [
    "dataset = SBU_dataset(SBU_dir)\n",
    "train, test = dataset.get_data(3) # slected from 0,1,2,3,4"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Initialize the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "os.environ[\"CUDA_DEVICE_ORDER\"]=\"PCI_BUS_ID\"   \n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"0\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "inp_j_0 (InputLayer)            (None, 16, 15, 3)    0                                            \n",
      "__________________________________________________________________________________________________\n",
      "inp_j_diff_0 (InputLayer)       (None, 16, 15, 3)    0                                            \n",
      "__________________________________________________________________________________________________\n",
      "inp_j_1 (InputLayer)            (None, 16, 15, 3)    0                                            \n",
      "__________________________________________________________________________________________________\n",
      "inp_j_diff_1 (InputLayer)       (None, 16, 15, 3)    0                                            \n",
      "__________________________________________________________________________________________________\n",
      "model_3 (Model)                 (None, 4, 4, 64)     11776       inp_j_0[0][0]                    \n",
      "                                                                 inp_j_diff_0[0][0]               \n",
      "                                                                 inp_j_1[0][0]                    \n",
      "                                                                 inp_j_diff_1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "maximum_2 (Maximum)             (None, 4, 4, 64)     0           model_3[1][0]                    \n",
      "                                                                 model_3[2][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "flatten_2 (Flatten)             (None, 1024)         0           maximum_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dropout_7 (Dropout)             (None, 1024)         0           flatten_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 256)          262400      dropout_7[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_18 (BatchNo (None, 256)          1024        dense_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "leaky_re_lu_18 (LeakyReLU)      (None, 256)          0           batch_normalization_18[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_8 (Dropout)             (None, 256)          0           leaky_re_lu_18[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dense_4 (Dense)                 (None, 8)            2056        dropout_8[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 277,256\n",
      "Trainable params: 276,296\n",
      "Non-trainable params: 960\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "lr=0.001\n",
    "adam = optimizers.Adam(lr)\n",
    "model = multi_obj()\n",
    "model.compile(adam, loss='mean_squared_error')\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1\n",
      "216/216 [==============================] - 3s 12ms/step - loss: 0.0011\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 984us/step - loss: 0.0012\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 903us/step - loss: 0.0013\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 790us/step - loss: 0.0013\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 775us/step - loss: 5.5415e-04\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 784us/step - loss: 0.0011\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 850us/step - loss: 0.0033\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 911us/step - loss: 0.0023\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 719us/step - loss: 0.0015\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 787us/step - loss: 0.0014\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 842us/step - loss: 0.0018\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 942us/step - loss: 0.0028\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 985us/step - loss: 6.5470e-04\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 669us/step - loss: 0.0022\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 788us/step - loss: 5.6142e-04\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 728us/step - loss: 0.0015\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 922us/step - loss: 3.7272e-04\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 1ms/step - loss: 0.0016\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 712us/step - loss: 0.0016\n",
      "Epoch 1/1\n",
      "216/216 [==============================] - 0s 752us/step - loss: 0.0010\n"
     ]
    }
   ],
   "source": [
    "epochs = 400\n",
    "for e in range(epochs):\n",
    "    X_0 = []\n",
    "    X_1 = []\n",
    "    X_2 = []\n",
    "    X_3 = []\n",
    "    Y = []\n",
    "\n",
    "    for i in range(1,9):                 # loop 8 classes\n",
    "        for j in range(len(train[i])):   # loop all samples within the same class\n",
    "            \n",
    "            #First person pose\n",
    "            p_0 = np.copy(train[i][j].T[:,:45])\n",
    "            p_0 = p_0.reshape([-1,15,3])\n",
    "            t_0 = p_0.shape[0]           # the number of all frames\n",
    "            if t_0>16:                   # sample the range from crop size of [16,t_0]\n",
    "                ratio = np.random.uniform(1,t_0/16)\n",
    "                l = int(16*ratio)\n",
    "                start = random.sample(range(t_0-l),1)[0]\n",
    "                end = start+l\n",
    "                p_0 = p_0[start:end,:,:]\n",
    "                p_0 = zoom(p_0)\n",
    "            elif t_0<16:\n",
    "                p_0 = zoom(p_0)\n",
    "\n",
    "            #Second person pose\n",
    "            p_1 = np.copy(train[i][j].T[:,45:])\n",
    "            p_1 = p_1.reshape([-1,15,3])\n",
    "            t_1 = p_1.shape[0]\n",
    "            if t_1 >16:  \n",
    "                ratio = np.random.uniform(1,t_1/16)\n",
    "                l = int(16*ratio)\n",
    "                start = random.sample(range(t_1-l),1)[0]\n",
    "                end = start+l\n",
    "                p_1 = p_1[start:end,:,:]\n",
    "                p_1 = zoom(p_1)\n",
    "            elif t_1 <16:\n",
    "                p_1 = zoom(p_1)\n",
    "            \n",
    "            # randomly mirror augmentation \n",
    "            # since two persions' postion could be switched\n",
    "            if np.random.choice([0,1],1): \n",
    "                p_0, p_1 = mirror(p_0,p_1)\n",
    "            \n",
    "            #Calculate the temporal difference\n",
    "            p_0_diff = p_0[1:,:,:]-p_0[:-1,:,:]\n",
    "            p_0_diff = np.concatenate((p_0_diff,np.expand_dims(p_0_diff[-1,:,:],axis=0)))\n",
    "            p_1_diff = p_1[1:,:,:]-p_1[:-1,:,:]\n",
    "            p_1_diff = np.concatenate((p_1_diff,np.expand_dims(p_1_diff[-1,:,:],axis=0)))\n",
    "\n",
    "            X_0.append(p_0)\n",
    "            X_1.append(p_0_diff)\n",
    "            X_2.append(p_1)\n",
    "            X_3.append(p_1_diff)\n",
    "\n",
    "            label = np.zeros(8)\n",
    "            label[i-1] = 1\n",
    "            Y.append(label)\n",
    "\n",
    "    X_0 = np.stack(X_0)\n",
    "    X_1 = np.stack(X_1)\n",
    "    X_2 = np.stack(X_2)\n",
    "    X_3 = np.stack(X_3)\n",
    "    Y = np.stack(Y)    \n",
    "\n",
    "    history = model.fit([X_0,X_1,X_2,X_3],Y,batch_size=32,epochs=1,verbose=True,shuffle=True)\n",
    "    \n",
    "    if not (e+1)%50:\n",
    "        lr *= 0.8\n",
    "        adam = optimizers.Adam(lr)\n",
    "        model.compile(adam, loss='mean_squared_error')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predict labels: [0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3\n",
      " 3 5 5 4 4 4 5 5 5 5 5 6 6 6 6 6 6 6 6 6 6 7 7 7 7 7 7 0 7]\n",
      "Ground truth labels: [0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3\n",
      " 3 3 3 4 4 4 5 5 5 5 5 6 6 6 6 6 6 6 6 6 6 7 7 7 7 7 7 7 7]\n"
     ]
    }
   ],
   "source": [
    "X_TEST_0 = []\n",
    "X_TEST_1 = []\n",
    "X_TEST_2 = []\n",
    "X_TEST_3 = []\n",
    "Y_TEST = []\n",
    "\n",
    "for i in range(1,9):\n",
    "    for j in range(len(test[i])):\n",
    "        p_0 = np.copy(test[i][j].T[:,:45])\n",
    "        p_0 = p_0.reshape([-1,15,3])\n",
    "        t_0 = p_0.shape[0]\n",
    "        # if the number of frame is more than 20, crop by scale 0.9, then rescale by interploration again\n",
    "        if t_0>=20: \n",
    "            p_0 = p_0[int(t_0*0.05):int(t_0*0.95),:,:]\n",
    "            p_0 = zoom(p_0)\n",
    "        elif t_0<20:\n",
    "            p_0 = zoom(p_0)\n",
    "        p_0_diff = p_0[1:,:,:]-p_0[:-1,:,:]\n",
    "        p_0_diff = np.concatenate((p_0_diff,np.expand_dims(p_0_diff[-1,:,:],axis=0)))\n",
    "        \n",
    "        p_1 = np.copy(test[i][j].T[:,45:])\n",
    "        p_1 = p_1.reshape([-1,15,3])\n",
    "        t_1 = p_1.shape[0]\n",
    "        if t_1 >=20:  \n",
    "            p_1 = p_1[int(t_1*0.05):int(t_1*0.95),:,:]\n",
    "            p_1 = zoom(p_1)\n",
    "        elif t_1 <20:\n",
    "            p_1 = zoom(p_1)\n",
    "        p_1_diff = p_1[1:,:,:]-p_1[:-1,:,:]\n",
    "        p_1_diff = np.concatenate((p_1_diff,np.expand_dims(p_1_diff[-1,:,:],axis=0)))\n",
    "\n",
    "        X_TEST_0.append(p_0)\n",
    "        X_TEST_1.append(p_0_diff)\n",
    "        X_TEST_2.append(p_1)\n",
    "        X_TEST_3.append(p_1_diff)\n",
    "        \n",
    "        label = np.zeros(8)\n",
    "        label[i-1] = 1\n",
    "        Y_TEST.append(label)\n",
    "\n",
    "X_TEST_0 = np.stack(X_TEST_0)\n",
    "X_TEST_1 = np.stack(X_TEST_1)\n",
    "X_TEST_2 = np.stack(X_TEST_2)\n",
    "X_TEST_3 = np.stack(X_TEST_3)\n",
    "Y_TEST = np.stack(Y_TEST)\n",
    "\n",
    "Y_pred = model.predict([X_TEST_0,X_TEST_1,X_TEST_2,X_TEST_3])\n",
    "\n",
    "print('Predict labels:',np.argmax(Y_pred,axis=1))\n",
    "print('Ground truth labels:',np.argmax(Y_TEST,axis=1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visulize the mirror augmentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/fan/anaconda3/envs/keras/lib/python3.6/site-packages/matplotlib/cbook/deprecation.py:107: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
      "  warnings.warn(message, mplDeprecation, stacklevel=1)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "i = 10\n",
    "draw_2d_pose([p_0[i],p_1[i]])\n",
    "p_0_new, p_1_new = mirror(p_0,p_1)\n",
    "draw_2d_pose([p_0_new[i],p_1_new[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}