SL_train.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import tensorflow as tf\n",
    "import keras\n",
    "import os, time, pickle\n",
    "from tqdm.notebook import tqdm\n",
    "\n",
    "from sl_model import SL512, DSODSL512\n",
    "from ssd_data import InputGenerator\n",
    "from sl_utils import PriorUtil\n",
    "from sl_training import SegLinkLoss, SegLinkFocalLoss\n",
    "\n",
    "from utils.model import load_weights, calc_memory_usage\n",
    "from utils.training import MetricUtility"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from data_synthtext import GTUtility\n",
    "\n",
    "file_name = 'gt_util_synthtext_seglink.pkl'\n",
    "with open(file_name, 'rb') as f:\n",
    "    gt_util = pickle.load(f)\n",
    "gt_util_train, gt_util_val = gt_util.split(0.9)\n",
    "\n",
    "print(gt_util_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# SegLink\n",
    "model = SL512()\n",
    "weights_path = './models/ssd512_voc_weights_fixed.hdf5'\n",
    "#weights_path = '~/.keras/models/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5'\n",
    "batch_size = 24\n",
    "experiment = 'sl512_synthtext'\n",
    "#experiment = 'sl512fl_synthtext'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# SegLink + DenseNet\n",
    "model = DSODSL512()\n",
    "#model = DSODSL512(activation='leaky_relu')\n",
    "weights_path = None\n",
    "batch_size = 6\n",
    "experiment = 'dsodsl512_synthtext'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# SegLink + ResNet\n",
    "from sl_model import SL512_resnet\n",
    "model = SL512_resnet()\n",
    "weights_path = None\n",
    "batch_size = 10\n",
    "experiment = 'sl512_resnet_synthtext'\n",
    "# NOTE: did not converge and I had no time for further investigations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "freeze = []\n",
    "\n",
    "if weights_path is not None:\n",
    "    if weights_path.find('ssd512') > -1:\n",
    "        layer_list = [\n",
    "            'conv1_1', 'conv1_2',\n",
    "            'conv2_1', 'conv2_2',\n",
    "            'conv3_1', 'conv3_2', 'conv3_3',\n",
    "            'conv4_1', 'conv4_2', 'conv4_3',\n",
    "            'conv5_1', 'conv5_2', 'conv5_3',\n",
    "            'fc6', 'fc7',\n",
    "            'conv6_1', 'conv6_2',\n",
    "            'conv7_1', 'conv7_2',\n",
    "            'conv8_1', 'conv8_2',\n",
    "            'conv9_1', 'conv9_2',\n",
    "        ]\n",
    "        freeze = [\n",
    "            'conv1_1', 'conv1_2',\n",
    "            'conv2_1', 'conv2_2',\n",
    "            'conv3_1', 'conv3_2', 'conv3_3',\n",
    "            #'conv4_1', 'conv4_2', 'conv4_3',\n",
    "            #'conv5_1', 'conv5_2', 'conv5_3',\n",
    "        ]\n",
    "        \n",
    "        load_weights(model, weights_path, layer_list)\n",
    "        for layer in model.layers:\n",
    "            layer.trainable = not layer.name in freeze\n",
    "    else:\n",
    "        load_weights(model, weights_path)\n",
    "\n",
    "prior_util = PriorUtil(model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "epochs = 100\n",
    "initial_epoch = 0\n",
    "\n",
    "#optimizer = tf.optimizers.SGD(learning_rate=1e-3, momentum=0.9, decay=0, nesterov=True)\n",
    "optimizer = tf.optimizers.Adam(learning_rate=1e-3, beta_1=0.9, beta_2=0.999, epsilon=0.001, decay=0.0)\n",
    "\n",
    "#loss = SegLinkLoss(lambda_offsets=1.0, lambda_links=1.0, neg_pos_ratio=3.0)\n",
    "loss = SegLinkFocalLoss(lambda_segments=100.0, lambda_offsets=1.0, lambda_links=100.0, gamma_segments=2, gamma_links=2)\n",
    "\n",
    "#regularizer = None\n",
    "regularizer = keras.regularizers.l2(5e-4) # None if disabled\n",
    "\n",
    "gen_train = InputGenerator(gt_util_train, prior_util, batch_size, model.image_size, augmentation=False)\n",
    "gen_val = InputGenerator(gt_util_val, prior_util, batch_size, model.image_size, augmentation=False)\n",
    "\n",
    "\n",
    "dataset_train, dataset_val = gen_train.get_dataset(), gen_val.get_dataset()\n",
    "iterator_train, iterator_val = iter(dataset_train), iter(dataset_val)\n",
    "\n",
    "checkdir = './checkpoints/' + time.strftime('%Y%m%d%H%M') + '_' + experiment\n",
    "\n",
    "if not os.path.exists(checkdir):\n",
    "    os.makedirs(checkdir)\n",
    "\n",
    "with open(checkdir+'/source.py','wb') as f:\n",
    "    source = ''.join(['# In[%i]\\n%s\\n\\n' % (i, In[i]) for i in range(len(In))])\n",
    "    f.write(source.encode())\n",
    "\n",
    "print(checkdir)\n",
    "\n",
    "for l in model.layers:\n",
    "    l.trainable = not l.name in freeze\n",
    "    if regularizer and l.__class__.__name__.startswith('Conv'):\n",
    "        model.add_loss(lambda l=l: regularizer(l.kernel))\n",
    "\n",
    "metric_util = MetricUtility(loss.metric_names, logdir=checkdir)\n",
    "\n",
    "@tf.function\n",
    "def step(x, y_true, training=False):\n",
    "    if training:\n",
    "        with tf.GradientTape() as tape:\n",
    "            y_pred = model(x, training=True)\n",
    "            metric_values = loss.compute(y_true, y_pred)\n",
    "            total_loss = metric_values['loss']\n",
    "            if len(model.losses):\n",
    "                total_loss += tf.add_n(model.losses)\n",
    "        gradients = tape.gradient(total_loss, model.trainable_variables)\n",
    "        optimizer.apply_gradients(zip(gradients, model.trainable_variables))\n",
    "    else:\n",
    "        y_pred = model(x, training=True)\n",
    "        metric_values = loss.compute(y_true, y_pred)\n",
    "    return metric_values\n",
    "\n",
    "#tf.profiler.experimental.start('./tblog')\n",
    "\n",
    "for k in tqdm(range(initial_epoch, epochs), 'total', leave=False):\n",
    "    print('\\nepoch %i/%i' % (k+1, epochs))\n",
    "    metric_util.on_epoch_begin()\n",
    "\n",
    "    for i in tqdm(range(gen_train.num_batches//4), 'training', leave=False):\n",
    "        x, y_true = next(iterator_train)\n",
    "        metric_values = step(x, y_true, training=True)\n",
    "        metric_util.update(metric_values, training=True)\n",
    "        #if i == 100: break\n",
    "    \n",
    "    model.save_weights(checkdir+'/weights.%03i.h5' % (k+1,))\n",
    "\n",
    "    for i in tqdm(range(gen_val.num_batches), 'validation', leave=False):\n",
    "        x, y_true = next(iterator_val)\n",
    "        metric_values = step(x, y_true, training=False)\n",
    "        metric_util.update(metric_values, training=False)\n",
    "        #if i == 10: break\n",
    "\n",
    "    metric_util.on_epoch_end(verbose=1)\n",
    "    #if k == 1: break\n",
    "\n",
    "#tf.profiler.experimental.stop()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": "3.8.10"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}