README.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "ocroline\n",
    "========\n",
    "\n",
    "An LSTM/CTC-based text line recognizer for printed text.\n",
    "\n",
    "In order to run this, you need a model. You can download a trained model\n",
    "for English from here:\n",
    "\n",
    "    https://storage.googleapis.com/tmb-models/line2-000003330-004377.pt\n",
    "\n",
    "Training is carried out using the `ocroline-train` command line\n",
    "program. This takes inputs in the form of tar files of training\n",
    "data as defined by [dlinputs](https://github.com/NVlabs/dlinputs).\n",
    "\n",
    "Recognition is carried out using the `ocroline.LineRecognizer` class.\n",
    "\n",
    "Note that `ocroline` requires a GPU for training, and the code currently also assumes a GPU for inference (although you could probably run the inference code without a GPU).\n",
    "\n",
    "There are some requirements for installation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-e git://github.com/NVlabs/dlinputs.git#egg=dlinputs\n",
      "-e git://github.com/NVlabs/dltrainers.git#egg=dltrainers\n",
      "-e git://github.com/tmbdev/cctc.git#egg=cctc\n",
      "editdistance\n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "cat requirements.txt"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Training files are just tar files containing `.png` and corresponding `.txt` files with the ground truth."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-rw-rw-r-- tmb/tmb           1 2018-03-24 23:57 000000.index\n",
      "-rw-rw-r-- tmb/tmb        1470 2018-03-24 23:57 000000.png\n",
      "-rw-rw-r-- tmb/tmb          19 2018-03-24 23:57 000000.txt\n",
      "-rw-rw-r-- tmb/tmb           1 2018-03-24 23:57 000001.index\n",
      "-rw-rw-r-- tmb/tmb         424 2018-03-24 23:57 000001.png\n",
      "-rw-rw-r-- tmb/tmb           3 2018-03-24 23:57 000001.txt\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "tar: write error\n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "tar -ztvf testdata/testlines.tgz | sed 6q"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is read internally using the `dlinputs` library. This library is also used for training and has a large number of options for conversions, batching, and data augmentation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    }
   ],
   "source": [
    "%pylab inline\n",
    "rc(\"image\", cmap=\"gray\", interpolation=\"bicubic\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Continuum Mechanics\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fea2b75e0d0>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from dlinputs import tarrecords\n",
    "sample = tarrecords.tariterator(open(\"testdata/testlines.tgz\")).next()\n",
    "print sample[\"txt\"]\n",
    "imshow(sample[\"png\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The training data for the line recognizer needs to be sized normalized. You can use the functions in `lineest.CenterLineNormalizer` for that. In this training set, the data is already normalized."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['CenterNormalizer', 'LineRecognizer', '__builtins__', '__doc__', '__file__', '__name__', '__package__', '__path__', 'lineest', 'recognizer']\n",
      "__key__ <type 'list'> ['000006', '000007', '000008', '000009', '000010']\n",
      "__source__ <type 'list'> ['testdata/testlines.tgz', 'testdata/testlines.tgz', 'testda\n",
      "_bucket 12\n",
      "image float64 (5, 48, 1295, 1)\n",
      "transcript float64 (5, 159, 97)\n",
      "input (3L, 48L, 20L, 1L)\n",
      "output (3L, 20L, 97L)\n",
      "Sequential(\n",
      "  (0): Reorder BHWD->BDHW\n",
      "  (1): CheckSizes [(0, 900), (1, 1), (48, 48), (0, 9000)]\n",
      "  (2): Conv2d(1, 100, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (3): ReLU()\n",
      "  (4): Reshape((0, [1, 2], 3))\n",
      "  (5): CheckSizes [(0, 900), (0, 5000), (0, 9000)]\n",
      "  (6): LSTM1:LSTM(4800, 100, bidirectional=1)\n",
      "  (7): Conv1d(200, 97, kernel_size=(1,), stride=(1,))\n",
      "  (8): Reorder BDW->BWD\n",
      "  (9): CheckSizes [(0, 900), (0, 9000), (97, 97)]\n",
      ")\n",
      "# 5\n",
      "TRU body fill a region in a four-dimensional Preisach space. A thermodynamical\n",
      "ALN \n",
      "PRE Z\n",
      "testset 100 1.0\n",
      "saving as testmodel-000000000-1000000.pt\n",
      "done\n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "ocroline-train -d testdata/testlines.tgz -t testdata/testlines.tgz -T 100 -o testmodel --epochs 1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note that `ocroline-train` uses the convention that all models are saved as `prefix-000000000-000000.pt`, where the first number is the number of training samples used for training (in thousands) and the second number is the error rate times one million."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Line Recognition"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%bash\n",
    "model=line2-000003330-004377.pt\n",
    "test -f $model || wget --quiet -nd https://storage.googleapis.com/tmb-models/$model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The `LineRecognizer` class wraps up the line recognizer in a small, simple-to-use class. This class will always perform line normalization on its input. It can also perform batching of multiple lines if you want to recognize multiple lines together."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): Reorder BHWD->BDHW\n",
       "  (1): CheckSizes [(0, 900), (1, 1), (48, 48), (0, 9000)]\n",
       "  (2): Conv2d(1, 100, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "  (3): BatchNorm2d(100, eps=1e-05, momentum=0.1, affine=True)\n",
       "  (4): ReLU()\n",
       "  (5): MaxPool2d(kernel_size=(2, 1), stride=(2, 1), dilation=(1, 1), ceil_mode=False)\n",
       "  (6): Conv2d(100, 200, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "  (7): BatchNorm2d(200, eps=1e-05, momentum=0.1, affine=True)\n",
       "  (8): ReLU()\n",
       "  (9): Reshape((0, [1, 2], 3))\n",
       "  (10): CheckSizes [(0, 900), (0, 5000), (0, 9000)]\n",
       "  (11): LSTM1:LSTM(4800, 200, bidirectional=1)\n",
       "  (12): Conv1d(400, 97, kernel_size=(1,), stride=(1,))\n",
       "  (13): Reorder BDW->BWD\n",
       "  (14): CheckSizes [(0, 900), (0, 9000), (97, 97)]\n",
       ")"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import ocroline\n",
    "rec = ocroline.LineRecognizer(\"line2-000003330-004377.pt\")\n",
    "rec.model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fe9ea75f290>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "imshow(sample[\"png\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Continuum Mechanics'"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rec.recognize_line(sample[\"png\"])"
   ]
  },
  {
   "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}