make tfRecods.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "IPython.notebook.set_autosave_interval(0)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Autosave disabled\n",
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    }
   ],
   "source": [
    "%autosave 0\n",
    "from __future__ import absolute_import, division, print_function\n",
    "%pylab inline\n",
    "\n",
    "\n",
    "import cv2\n",
    "import scipy.misc\n",
    "import skimage.util\n",
    "\n",
    "import numpy as np\n",
    "import glob\n",
    "import os.path\n",
    "import xml.etree.ElementTree as ET\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import scipy.io\n",
    "import random\n",
    "from scipy import ndimage\n",
    "from skimage import io\n",
    "from skimage import transform\n",
    "import math\n",
    "\n",
    "from PIL import Image, ImageEnhance\n",
    "from scipy import signal\n",
    "from sklearn import preprocessing\n",
    "from sklearn.decomposition import PCA\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'': 28,\n",
       " ' ': 27,\n",
       " '%%': 0,\n",
       " 'a': 1,\n",
       " 'b': 2,\n",
       " 'c': 3,\n",
       " 'd': 4,\n",
       " 'e': 5,\n",
       " 'f': 6,\n",
       " 'g': 7,\n",
       " 'h': 8,\n",
       " 'i': 9,\n",
       " 'j': 10,\n",
       " 'k': 11,\n",
       " 'l': 12,\n",
       " 'm': 13,\n",
       " 'n': 14,\n",
       " 'o': 15,\n",
       " 'p': 16,\n",
       " 'q': 17,\n",
       " 'r': 18,\n",
       " 's': 19,\n",
       " 't': 20,\n",
       " 'u': 21,\n",
       " 'v': 22,\n",
       " 'w': 23,\n",
       " 'x': 24,\n",
       " 'y': 25,\n",
       " 'z': 26}"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vocabulary = {}\n",
    "vocabulary_path = \"some.mat\"\n",
    "def getIndex(c,voc):\n",
    "    for name, age in voc.iteritems():\n",
    "        if age == c:\n",
    "            return name\n",
    "    print(\"-\"*30,\"error-\",c)\n",
    "    return None\n",
    "\n",
    "if os.path.exists(vocabulary_path):\n",
    "    imgs=scipy.io.loadmat(vocabulary_path)['vocabulary']\n",
    "else:\n",
    "    nrC=1\n",
    "    vocabulary['%%'] = 0 \n",
    "    '''\n",
    "    c = '0'\n",
    "    \n",
    "    while ord(c) != ord('9')+1:\n",
    "        vocabulary[c] = nrC\n",
    "        nrC = nrC + 1\n",
    "        c = chr(ord(c)+1)\n",
    "    c = 'A'\n",
    "    while ord(c) != ord('Z')+1:\n",
    "        vocabulary[c] = nrC\n",
    "        nrC = nrC + 1\n",
    "        c = chr(ord(c)+1)\n",
    "    '''\n",
    "    c = 'a'\n",
    "    while ord(c) != ord('z')+1:\n",
    "        vocabulary[c] = nrC\n",
    "        nrC = nrC + 1\n",
    "        c = chr(ord(c)+1)\n",
    "    '''\n",
    "    cr = [',','.','\"','\\'',' ','-','#','(',')',';','?',':','*','&','!','/','']\n",
    "    for c in cr:\n",
    "        vocabulary[c] = nrC\n",
    "        nrC = nrC + 1\n",
    "    '''\n",
    "    vocabulary[' '] = nrC\n",
    "    nrC += 1\n",
    "    vocabulary[''] = nrC\n",
    "    \n",
    "def preprocessTarget(line):\n",
    "    \n",
    "    QUOT_TOKEN_Init = \"&quot;\"\n",
    "    QUOT_TOKEN = '\"'\n",
    "           \n",
    "    FIRST_INDEX = ord('a') - 4  # 0 is reserved to space\n",
    "\n",
    "    targets = line.strip().replace(QUOT_TOKEN_Init, QUOT_TOKEN)\n",
    "        \n",
    "    targets = np.asarray([vocabulary[x]  for x in targets])\n",
    "    \n",
    "    return targets\n",
    "\n",
    "\n",
    "\n",
    "vocabulary "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3542, 2479)\n",
      "17\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0, 0)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fb08df45450>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "maxL = 100#inaltimea\n",
    "maxi=0\n",
    "def processPage(nameXml,nameImg,words=False,maxW=400):\n",
    "    tree = ET.parse(nameXml)\n",
    "    root = tree.getroot()\n",
    "    global maxi\n",
    "    letters = list()\n",
    "    coords = list()\n",
    "    form = cv2.imread(nameImg, cv2.IMREAD_GRAYSCALE )\n",
    "    print(form.shape)\n",
    "    imgs = []\n",
    "    texts = []\n",
    "    lengths = []\n",
    "    for i,line in enumerate(root[1]):\n",
    "        #plt.subplot(10,1,i+1)\n",
    "        if int(line.attrib['dsy'])-int(line.attrib['asy'])>maxL:\n",
    "            continue\n",
    "        if words: \n",
    "            b = 0\n",
    "            e = 0\n",
    "            t = \"\"\n",
    "            ss=0\n",
    "            jj=0\n",
    "            for w in line:                \n",
    "                if(w.tag == \"lower-contour\" or w.tag == \"upper-contour\" or len(w) == 0):\n",
    "                        continue\n",
    "                #print (w.attrib)\n",
    "                \n",
    "                b1 = 0\n",
    "                e1 = 0\n",
    "                sss=0\n",
    "                jjj=0\n",
    "                for letter in w:\n",
    "                    if b1==0:\n",
    "                        b1 = int(letter.attrib['x'])\n",
    "                        jjj = int(letter.attrib['y'])\n",
    "                    e1 = max(e1,int(letter.attrib['x'])+int(letter.attrib['width']))\n",
    "                    b1 = min(b1,int(letter.attrib['x']))\n",
    "                    jjj = min(jjj,int(letter.attrib['y']))\n",
    "                    sss = max(sss,int(letter.attrib['y'])+int(letter.attrib['height']))\n",
    "                \n",
    "                #b1 = int(w[0].attrib['x'])\n",
    "                #e1 = int(w[len(w)-1].attrib['x']) + int(w[len(w)-1].attrib['width'])\n",
    "                if b==0:\n",
    "                    b=b1\n",
    "                    e=e1\n",
    "                    ss=sss\n",
    "                    jj=jjj\n",
    "                if e1-b>maxW and t!=\"\":                        \n",
    "                        img = form[jj:ss,b:e]\n",
    "                        img = cv2.resize(img, (0,0), fx=0.3, fy=0.3)\n",
    "                        ret,thresh1 = cv2.threshold(img,int(line.attrib['threshold']),255,cv2.THRESH_BINARY)\n",
    "                        thresh1 =1.-(thresh1/255)\n",
    "                        ok=1\n",
    "                        for llt in t:\n",
    "                            if llt not in vocabulary:\n",
    "                                ok = 0\n",
    "                                break\n",
    "                        if ok==1  and len(t)*4<thresh1.shape[1] and ss-jj<maxL:\n",
    "                            maxi=max(maxi,img.shape[0])\n",
    "                            texts.append(preprocessTarget(t))\n",
    "                            imgs.append(img)\n",
    "                            lengths.append(thresh1.shape[1])\n",
    "                            #print(t)\n",
    "                            if t.strip()==\"it is\" and False :\n",
    "                                print (nameXml)\n",
    "                                plt.imshow(thresh1)\n",
    "                                return None\n",
    "                        if e1-b1 > maxW:\n",
    "                            b=0\n",
    "                            t=\"\"\n",
    "                            continue\n",
    "                        b = b1\n",
    "                        e = e1\n",
    "                        jj = min(jj,jjj)\n",
    "                        ss= max(ss,sss)\n",
    "                        t=w.attrib['text']\n",
    "                else:\n",
    "                        if e1-b1 > maxW:\n",
    "                            b=0\n",
    "                            continue\n",
    "                        t = t + \" \" + w.attrib['text']\n",
    "                        e = e1\n",
    "                        jj = min(jj,jjj)\n",
    "                        ss= max(ss,sss)\n",
    "                    #print (letter.attrib)\n",
    "            if e-b < maxW and t!=\"\":\n",
    "                    #correct latimea\n",
    "                    img = form[jj:ss,b:e]#int(line.attrib['asy']):int(line.attrib['dsy'])\n",
    "                    img = cv2.resize(img, (0,0), fx=0.3, fy=0.3)\n",
    "                    ret,thresh1 = cv2.threshold(img,int(line.attrib['threshold']),255,cv2.THRESH_BINARY)                    \n",
    "                    maxi=max(maxi,img.shape[0])\n",
    "                    ok = 1\n",
    "                    for llt in t:\n",
    "                            if llt not in vocabulary:\n",
    "                                ok = 0\n",
    "                                break\n",
    "                    if ok==1 and len(t)*4<thresh1.shape[1] and ss-jj<maxL:\n",
    "                        texts.append(preprocessTarget(t))\n",
    "                        thresh1 =1.-(thresh1/255)\n",
    "                        imgs.append(img)\n",
    "                        lengths.append(thresh1.shape[1])\n",
    "                        #print(t)\n",
    "                        if t.strip()==\"not in\" and False:\n",
    "                                plt.imshow(thresh1)\n",
    "                                #print(t)\n",
    "                                print(line.attrib)\n",
    "                                print (nameXml)\n",
    "                                return None\n",
    "            #break    \n",
    "        else:\n",
    "            print(11)\n",
    "            img = form[int(line.attrib['asy']):int(line.attrib['dsy'])]\n",
    "            ret,thresh1 = cv2.threshold(img,int(line.attrib['threshold']),255,cv2.THRESH_BINARY)\n",
    "            \n",
    "            maxi=max(maxi,img.shape[0])\n",
    "            texts.append(preprocessTarget(line.attrib['text']))\n",
    "            imgs.append(thresh1)\n",
    "            lengths.append(thresh1.shape[1])\n",
    "        \n",
    "    return [texts,imgs,lengths]\n",
    "\n",
    "    \n",
    "nameXml = '../xml/b05-074.xml'\n",
    "nameImg = '../forms/b05-074.png'#'forms/a01-030u.png'#'forms/a01-113.png'\n",
    "\n",
    "t=processPage(nameXml,nameImg,True)\n",
    "#t[0]\n",
    "#print (maxi)\n",
    "print (len(t[0]))\n",
    "#for e,i in enumerate(t[0]):\n",
    "#    print(decodePrediction([[],t[0][e]]))\n",
    "#plt.imshow(t[1][0])\n",
    "l = 0;h = 0\n",
    "plt.figure(figsize=(15,25))\n",
    "for e in range(len(t[0][-9*2:])):\n",
    "    break\n",
    "    l = max(l,len(t[1][e][:,0]))\n",
    "    h = max(h,len(t[1][e][0]))\n",
    "    plt.subplot(9,2,e+1)\n",
    "    plt.imshow(t[1][e])\n",
    "l,h"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fb0ec0c3b10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "def get_features(img, thesh=100):\n",
    "    l = [img]\n",
    "    for i in range(3):\n",
    "        l.append(cv2.Sobel(img,cv2.CV_64F,1,0,ksize=(i*2+1)))\n",
    "        l.append(cv2.Sobel(img,cv2.CV_64F,0,1,ksize=(i*2+1)))\n",
    "    for i in range(1):\n",
    "        l.append(cv2.Canny(img,100+10*i,200))\n",
    "    for i in range(3):#block size, ksize, k harris\n",
    "        for j in range(3):\n",
    "                l.append(cv2.cornerHarris(img,2+j,(i*2+1),5))\n",
    "    theta, bwimg = cv2.threshold(img, 100, 255, cv2.THRESH_OTSU)\n",
    "    l.append(bwimg)\n",
    "    #l.append(correctSlant(img,tt=False))\n",
    "    l = [(i-np.amin(i))/(np.amax(i)-np.amin(i)) for i in l]\n",
    "    return l\n",
    "    #p = [np.expand_dims(preprocessing.scale(np.float32(i)),2) for i in l]\n",
    "    #return np.concatenate(p,2)\n",
    "imgs = get_features(t[1][0])\n",
    "#print(imgs.shape)\n",
    "plt.figure(figsize=(15,25))\n",
    "for i  in range(len(imgs)):\n",
    "    break\n",
    "    plt.subplot(20,2,i+1),plt.imshow(imgs[i],cmap = 'gray')\n",
    "    plt.title('{}'.format(i)), plt.xticks([]), plt.yticks([]), colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(24, 36, 24, 18) (28, 112)\n",
      "[17, 21, 19, 11, 9, 24, 22, 20, 24, 24, 24, 24, 24, 24, 19, 1, 17]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.colorbar.Colorbar instance at 0x7fb0c4af2170>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAOkAAAEACAYAAABWC1EdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ncZX8/2Eu8Czs/vytAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fb0908a4590>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def getSlices(img,w,stride=3,W=130,H=36):\n",
    "    img = 1.0 - img\n",
    "    assert w%2==0\n",
    "    w1 = img.shape[1]\n",
    "    ll = 100\n",
    "    l = []\n",
    "    #imgr = resize(img,height=H)\n",
    "    #if imgr.shape[1]<=W:\n",
    "    #    img = imgr\n",
    "    if W!=None and H!=None:\n",
    "        if img.shape[0]!=H and int((H-img.shape[0])/2)>0:\n",
    "            img = np.insert(img, [0]*int((H-img.shape[0])/2), 0.5, axis=0)\n",
    "        if img.shape[0]!=H:\n",
    "            img = np.insert(img, [img.shape[0]]*(H-img.shape[0]), 0.5, axis=0)\n",
    "        if img.shape[1]!=W:\n",
    "            img = np.insert(img, [img.shape[1]]*(W-img.shape[1]), 0.5, axis=1)    \n",
    "    if img.shape[1]%w!=0 and False:\n",
    "        img = np.insert(img, [img.shape[1]]*(w-(img.shape[1]%w)), 0.5, axis=1)\n",
    "    i = 0\n",
    "    #return img, w1\n",
    "    #add begining and ending offset\n",
    "    #img = np.insert(img, [img.shape[1]]*(int(w*0.6)), 0, axis=1)  \n",
    "    #img = np.insert(img, [0]*(int(w*0.6)), 0, axis=1)  \n",
    "    \n",
    "    #make slices\n",
    "    \n",
    "    cosineWindow = [signal.cosine(w) for i in range(H)]\n",
    "    while i<img.shape[1]-w:\n",
    "        if i>w1 and ll == 100:\n",
    "            ll = len(l)\n",
    "        #apply cosine transform\n",
    "        frame = img[:,i:i+w]\n",
    "        #frame = frame*np.array(cosineWindow)      \n",
    "        l.append(frame)\n",
    "        i=i+3\n",
    "    if len(np.asarray(l)) < ll+1:\n",
    "        ll = len(np.asarray(l))-1\n",
    "    return np.asarray(l), ll+1#len is not correct!!\n",
    "\n",
    "def getSlicesImgs(imgs,w):\n",
    "    nImgs = []\n",
    "    nln = []\n",
    "    for img in imgs:\n",
    "        #correct slant\n",
    "        #aa = findAngle(img,thesh=100)\n",
    "        #print(np.amin(img))\n",
    "        #print(aa)\n",
    "        #img = shear(img,aa,binar=False)\n",
    "        #print(np.amin(img))\n",
    "        #get filters\n",
    "        imgT = get_features(img)\n",
    "        #print(imgT[0].shape,len(imgT))\n",
    "        \n",
    "        ims = []\n",
    "        for ii in imgT:\n",
    "            im, ln = getSlices(ii,w)\n",
    "            ims.append(im*-1.)\n",
    "        #print(ims[0].shape,len(ims))\n",
    "        \n",
    "        nln.append(ln)\n",
    "        ims = [np.expand_dims(np.float32(i),3) for i in ims]\n",
    "        ims = np.concatenate(ims,3)\n",
    "        #print(ims.shape)\n",
    "        nImgs.append(ims)\n",
    "        #return 0\n",
    "    return nImgs, nln\n",
    "\n",
    "i,k = getSlicesImgs(t[1],24)\n",
    "print(i[1].shape,t[1][5].shape)\n",
    "print(k)\n",
    "plt.imshow(i[0][3,:,:,0]);plt.colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "../forms/a01-000u.png\n",
      "(3542, 2479)\n",
      "../forms/a01-000x.png\n",
      "(3542, 2479)\n",
      "../forms/a01-003.png\n",
      "(3542, 2479)\n",
      "../forms/a01-003u.png\n",
      "(3542, 2479)\n",
      "0 -l 0\n",
      "(25, 350) **********\n",
      "error!! (27, 52)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (28, 96)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 30)\n",
      "(25, 350) **********\n",
      "error!! (26, 50)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 104)\n",
      "(25, 350) **********\n",
      "error!! (26, 59)\n",
      "(25, 350) **********\n",
      "error!! (26, 70)\n",
      "(25, 350) **********\n",
      "error!! (26, 50)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "1 the file saved! test-batch1/handwritten-test-1.tfrecords\n",
      "../forms/a01-003x.png\n",
      "(3542, 2479)\n",
      "../forms/a01-007.png\n",
      "(3542, 2479)\n",
      "../forms/a01-007u.png\n",
      "(3542, 2479)\n",
      "1 -l 0\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 46)\n",
      "(25, 350) **********\n",
      "error!! (27, 64)\n",
      "(25, 350) **********\n",
      "error!! (29, 44)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (28, 46)\n",
      "(25, 350) **********\n",
      "error!! (28, 55)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "2 the file saved! test-batch1/handwritten-test-2.tfrecords\n",
      "../forms/a01-007x.png\n",
      "(3542, 2479)\n",
      "2 -l 0\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 86)\n",
      "(25, 350) **********\n",
      "error!! (27, 45)\n",
      "(25, 350) **********\n",
      "error!! (27, 103)\n",
      "(25, 350) **********\n",
      "error!! (28, 79)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (27, 48)\n",
      "(25, 350) **********\n",
      "error!! (27, 59)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (27, 113)\n",
      "(25, 350) **********\n",
      "3 the file saved! test-batch1/handwritten-test-3.tfrecords\n",
      "../forms/a01-011.png\n",
      "(3542, 2479)\n",
      "../forms/a01-011u.png\n",
      "(3542, 2479)\n",
      "../forms/a01-011x.png\n",
      "(3542, 2479)\n",
      "../forms/a01-014.png\n",
      "(3542, 2479)\n",
      "../forms/a01-014u.png\n",
      "(3542, 2479)\n",
      "3 -l 0\n",
      "(25, 350) **********\n",
      "error!! (28, 101)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (27, 23)\n",
      "(25, 350) **********\n",
      "error!! (27, 83)\n",
      "(25, 350) **********\n",
      "error!! (27, 101)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (27, 74)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "4 the file saved! test-batch1/handwritten-test-4.tfrecords\n",
      "../forms/a01-014x.png\n",
      "(3542, 2479)\n",
      "../forms/a01-020.png\n",
      "(3542, 2479)\n",
      "4 -l 0\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (27, 108)\n",
      "(25, 350) **********\n",
      "error!! (27, 37)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 94)\n",
      "(25, 350) **********\n",
      "error!! (26, 23)\n",
      "(25, 350) **********\n",
      "error!! (26, 117)\n",
      "(25, 350) **********\n",
      "error!! (26, 52)\n",
      "(25, 350) **********\n",
      "error!! (28, 114)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 63)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "5 the file saved! test-batch1/handwritten-test-5.tfrecords\n",
      "../forms/a01-020u.png\n",
      "(3542, 2479)\n",
      "5 -l 0\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 88)\n",
      "(25, 350) **********\n",
      "error!! (26, 63)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "6 the file saved! test-batch1/handwritten-test-6.tfrecords\n",
      "../forms/a01-020x.png\n",
      "(3542, 2479)\n",
      "6 -l 0\n",
      "(25, 350) **********\n",
      "error!! (28, 118)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (28, 113)\n",
      "(25, 350) **********\n",
      "error!! (28, 101)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 37)\n",
      "(25, 350) **********\n",
      "error!! (26, 117)\n",
      "(25, 350) **********\n",
      "error!! (28, 100)\n",
      "(25, 350) **********\n",
      "error!! (27, 94)\n",
      "(25, 350) **********\n",
      "7 the file saved! test-batch1/handwritten-test-7.tfrecords\n",
      "../forms/a01-026.png\n",
      "(3542, 2479)\n",
      "../forms/a01-026u.png\n",
      "(3542, 2479)\n",
      "7 -l 0\n",
      "error!! (27, 34)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 104)\n",
      "(25, 350) **********\n",
      "error!! (26, 98)\n",
      "(25, 350) **********\n",
      "error!! (26, 93)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (26, 117)\n",
      "(25, 350) **********\n",
      "error!! (28, 98)\n",
      "(25, 350) **********\n",
      "error!! (28, 36)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "error!! (29, 91)\n",
      "(25, 350) **********\n",
      "error!! (29, 114)\n",
      "(25, 350) **********\n",
      "error!! (29, 44)\n",
      "(25, 350) **********\n",
      "error!! (28, 70)\n",
      "(25, 350) **********\n",
      "error!! (28, 89)\n",
      "(25, 350) **********\n",
      "error!! (28, 39)\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "(25, 350) **********\n",
      "8 the file saved! test-batch1/handwritten-test-8.tfrecords\n",
      "../forms/a01-026x.png\n",
      "(3542, 2479)\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from IPython.display import clear_output\n",
    "\n",
    "def addPaddingTF(imgs):\n",
    "    imgs = [imgs]\n",
    "    at = []\n",
    "    H=25#120#347\n",
    "    W = 350#300\n",
    "    for k,img1 in enumerate(imgs):\n",
    "        img = imgs[k]\n",
    "        if img.shape[0]!=H and int((H-img.shape[0])/2)>0:\n",
    "            img = np.insert(img, [0]*int((H-img.shape[0])/2), 0, axis=0)\n",
    "        if img.shape[0]!=H:\n",
    "            img = np.insert(img, [img.shape[0]]*(H-img.shape[0]), 0, axis=0)\n",
    "        if img.shape[1]!=W:\n",
    "            img = np.insert(img, [img.shape[1]]*(W-img.shape[1]), 0, axis=1)\n",
    "        at.append(img)\n",
    "    return at[0]\n",
    "\n",
    "def makeBatchesTFRecord(pathSave,pathXML,pathImgs,batchSize,words=False):\n",
    "    if not os.path.exists(pathSave):\n",
    "        os.makedirs(pathSave)\n",
    "    totalT, totalI, totalL = [],[],[]\n",
    "    nr=0\n",
    "    maxle=0\n",
    "    txtLD = {}\n",
    "    dd,ll=0,0\n",
    "        \n",
    "    for xml in sorted(glob.glob(pathXML+\"*.xml\"))[:20]:#restrict number of images processed, edit it!!!\n",
    "            name = xml.split('.xml')[0].split('/')[-1]\n",
    "            img = pathImgs+name+'.png'\n",
    "            print(img)\n",
    "            if img == \"forms/g06-011p.png\":\n",
    "                break\n",
    "            if not os.path.isfile(img) :\n",
    "                continue\n",
    "            \n",
    "            texts,imgs,lengths = processPage(xml,img,words)\n",
    "                      \n",
    "            #imgs,lengths = getSlicesImgs(imgs,12)#put the number of colums in a slice: edit it\n",
    "            \n",
    "            totalT.extend(texts)\n",
    "            totalI.extend(imgs)\n",
    "            totalL.extend(lengths)\n",
    "            \n",
    "            if(len(totalT)>=batchSize):\n",
    "                #save batch\n",
    "                print(nr,'-l',maxle)\n",
    "                if True:\n",
    "                    writer = tf.python_io.TFRecordWriter(pathSave+\"/handwritten-test-{}.tfrecords\".format(nr))\n",
    "                    for ii in range(batchSize):\n",
    "                        if totalI[ii].shape[0] >25  or \\\n",
    "                        totalI[ii].shape[1]>350 :\n",
    "                            print('error!!', totalI[ii].shape)\n",
    "                            totalI[ii] = cv2.resize(totalI[ii],(350,25))\n",
    "                            #continue\n",
    "                        imgI = addPaddingTF(totalI[ii])  \n",
    "                        print(imgI.shape,'*'*10)\n",
    "                        # construct the Example proto boject\n",
    "                        example = tf.train.Example(\n",
    "                            # Example contains a Features proto object\n",
    "                            features=tf.train.Features(\n",
    "                              # Features contains a map of string to Feature proto objects\n",
    "                              feature={\n",
    "                                # A Feature contains one of either a int64_list,\n",
    "                                # float_list, or bytes_list\n",
    "                                'seq_len': tf.train.Feature(\n",
    "                                    int64_list=tf.train.Int64List(value=[totalL[ii]])),\n",
    "                                'target': tf.train.Feature(\n",
    "                                    int64_list=tf.train.Int64List(value=totalT[ii].astype(\"int64\"))),\n",
    "                                'imageInput': tf.train.Feature(\n",
    "                                    float_list=tf.train.FloatList(value=(imgI).reshape(-1).astype(\"float\"))),\n",
    "                        }))\n",
    "                        # use the proto object to serialize the example to a string\n",
    "                        serialized = example.SerializeToString()\n",
    "                        # write the serialized object to disk\n",
    "                        writer.write(serialized)\n",
    "                    writer.close()\n",
    "                nr += 1  \n",
    "                print(nr,'the file saved!',pathSave+\"/handwritten-test-{}.tfrecords\".format(nr))\n",
    "                #clear_output()\n",
    "                totalT = totalT[batchSize:]\n",
    "                totalI = totalI[batchSize:]\n",
    "                totalL = totalL[batchSize:]\n",
    "    target = open(pathSave+'readme.txt', 'w')\n",
    "    target.write(\"max len text:\"+str(maxle)+\"\\n\")\n",
    "    target.write(\"image shape:50x300 \\n\")\n",
    "    target.close()\n",
    "    \n",
    "makeBatchesTFRecord('test-batch1','../xml/','../forms/',20,True)\n",
    "#"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "python train.py --slices 55 --width 12 --stride 1  --Bwidth 350 --vocabulary_size 29 \\\n",
    "--height 25 --train_data_pattern test-batch1/handwritten-test-{}.tfrecords --train_dir models-feds \\\n",
    "--test_data_pattern test-batch1/handwritten-test-{}.tfrecords  --max_steps 20 --batch_size 20 --beam_size 1 \\\n",
    "--input_chanels 1 --start_new_model --rnn_cell LSTM --model LSTMCTCModel --num_epochs 6000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    " python inference.py --slices 55 --width 12 --Bwidth 350 --stride 1 \\\n",
    "    --input_chanels 1 --height 25 --input_data_pattern  test-batch1/handwritten-test-1.tfrecords \\\n",
    "    --train_dir models-feds  --batch_size 20 --beam_size 1 "
   ]
  }
 ],
 "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": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}