final_training_trans.ipynb

{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "G3uernfiXfzU",
        "outputId": "43a48fd6-020b-4987-8642-96555a345e9a"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "SRC: cloth3d++_subset/\n",
            "Number of data used to train:\n",
            "59928\n",
            "Input shape: (2, 256, 256, 3)\n",
            "Label shape: (2, 256, 256)\n",
            "Learning rate: 1e-07\n",
            "Epoch 1/12\n",
            "  14982/Unknown - 2378s 157ms/step - loss: 0.0059 - mae: 0.0394"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "2024-05-27 00:55:51.346384: I tensorflow/core/framework/local_rendezvous.cc:405] Local rendezvous recv item cancelled. Key hash: 16569504095248563035\n",
            "2024-05-27 00:55:51.346424: I tensorflow/core/framework/local_rendezvous.cc:405] Local rendezvous recv item cancelled. Key hash: 2548887854972656661\n",
            "2024-05-27 00:55:51.346431: I tensorflow/core/framework/local_rendezvous.cc:405] Local rendezvous recv item cancelled. Key hash: 12272699694009900839\n",
            "/home/alberto/anaconda3/envs/tf3dcloth/lib/python3.11/site-packages/keras/src/engine/training.py:3000: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')`.\n",
            "  saving_api.save_model(\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "14982/14982 [==============================] - 2489s 165ms/step - loss: 0.0059 - mae: 0.0394 - val_loss: 0.0061 - val_mae: 0.0288\n",
            "Epoch 2/12\n",
            "14982/14982 [==============================] - 2470s 165ms/step - loss: 0.0026 - mae: 0.0191 - val_loss: 0.0059 - val_mae: 0.0263\n",
            "Epoch 3/12\n",
            "14982/14982 [==============================] - 2472s 165ms/step - loss: 0.0021 - mae: 0.0163 - val_loss: 0.0054 - val_mae: 0.0251\n",
            "Epoch 4/12\n",
            "14982/14982 [==============================] - 2470s 165ms/step - loss: 0.0018 - mae: 0.0146 - val_loss: 0.0058 - val_mae: 0.0254\n",
            "Epoch 5/12\n",
            "14982/14982 [==============================] - 2473s 165ms/step - loss: 0.0016 - mae: 0.0135 - val_loss: 0.0057 - val_mae: 0.0246\n",
            "Epoch 6/12\n",
            "14982/14982 [==============================] - 2474s 165ms/step - loss: 0.0014 - mae: 0.0127 - val_loss: 0.0050 - val_mae: 0.0229\n",
            "Epoch 7/12\n",
            "14982/14982 [==============================] - 2474s 165ms/step - loss: 0.0013 - mae: 0.0120 - val_loss: 0.0048 - val_mae: 0.0223\n",
            "Epoch 8/12\n",
            "14982/14982 [==============================] - 2475s 165ms/step - loss: 0.0012 - mae: 0.0114 - val_loss: 0.0047 - val_mae: 0.0222\n",
            "Epoch 9/12\n",
            "14982/14982 [==============================] - 2473s 165ms/step - loss: 0.0011 - mae: 0.0109 - val_loss: 0.0049 - val_mae: 0.0223\n",
            "Epoch 10/12\n",
            "14982/14982 [==============================] - 2472s 165ms/step - loss: 0.0010 - mae: 0.0104 - val_loss: 0.0046 - val_mae: 0.0218\n",
            "Epoch 11/12\n",
            "14982/14982 [==============================] - 2473s 165ms/step - loss: 9.9547e-04 - mae: 0.0101 - val_loss: 0.0047 - val_mae: 0.0216\n",
            "Epoch 12/12\n",
            "14982/14982 [==============================] - 2473s 165ms/step - loss: 9.3181e-04 - mae: 0.0097 - val_loss: 0.0045 - val_mae: 0.0209\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 1500x400 with 2 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Best validation loss: 0.004457799252122641 at epoch 11\n"
          ]
        }
      ],
      "source": [
        "\n",
        "import os\n",
        "import glob\n",
        "import pickle\n",
        "import numpy as np\n",
        "from PIL import Image\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "from scipy.ndimage import binary_dilation, binary_erosion\n",
        "from skimage.transform import resize\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "from keras_unet_collection import models\n",
        "from tensorflow import keras\n",
        "from keras_unet_collection.utils import dummy_loader\n",
        "import cv2\n",
        "import gc\n",
        "\n",
        "import os\n",
        "import glob\n",
        "import numpy as np\n",
        "from PIL import Image\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "from scipy.ndimage import binary_dilation, binary_erosion\n",
        "from skimage.transform import resize\n",
        "import tensorflow as tf\n",
        "import os\n",
        "from PIL import Image\n",
        "import tensorflow_addons as tfa\n",
        "\n",
        "\n",
        "\n",
        "# Path to data\n",
        "SRC = 'cloth3d++_subset/'\n",
        "print('SRC:', SRC)\n",
        "\n",
        "n_epochs = 20\n",
        "batch_size = 2\n",
        "\n",
        "# %%\n",
        "class TFRecordDataHandler:\n",
        "    def __init__(self, tfrecord_file, batch_size=32, shuffle=True, augment=False):\n",
        "        self.tfrecord_file = tfrecord_file\n",
        "        self.batch_size = batch_size\n",
        "        self.shuffle = shuffle\n",
        "        self.augment = augment\n",
        "\n",
        "    def _parse_function(self, proto):\n",
        "        feature_description = {\n",
        "            'image': tf.io.FixedLenFeature([], tf.string),\n",
        "            'depth': tf.io.FixedLenFeature([], tf.string),\n",
        "            'height': tf.io.FixedLenFeature([], tf.int64),\n",
        "            'width': tf.io.FixedLenFeature([], tf.int64),\n",
        "            'depth_height': tf.io.FixedLenFeature([], tf.int64),\n",
        "            'depth_width': tf.io.FixedLenFeature([], tf.int64)\n",
        "        }\n",
        "        parsed_features = tf.io.parse_single_example(proto, feature_description)\n",
        "\n",
        "        image = tf.io.decode_raw(parsed_features['image'], tf.uint8)\n",
        "        depth = tf.io.decode_raw(parsed_features['depth'], tf.float32)\n",
        "\n",
        "        height = parsed_features['height']\n",
        "        width = parsed_features['width']\n",
        "\n",
        "        image = tf.reshape(image, [height, width, 3])\n",
        "        depth = tf.reshape(depth, [parsed_features['depth_height'], parsed_features['depth_width']])\n",
        "\n",
        "        return image, depth\n",
        "\n",
        "    def _normalize(self, image, depth):\n",
        "        # Convert image to float for processing and normalize to range [0, 1]\n",
        "        image = tf.cast(image, tf.float32) / 255.0\n",
        "\n",
        "        # Create a mask where depth values are greater than zero\n",
        "        depth_mask = depth > 0\n",
        "\n",
        "        # Normalize depth based on masked regions\n",
        "        # Calculate the mean of the depth where it is greater than zero\n",
        "        depth_values = tf.boolean_mask(depth, depth_mask)\n",
        "        depth_mean = tf.reduce_mean(depth_values)\n",
        "\n",
        "        # Subtract the mean from the depth values where mask is true\n",
        "        depth = tf.where(depth_mask, depth - depth_mean, depth)\n",
        "\n",
        "        # Set depth values less than 0 to 0 after subtraction\n",
        "        depth = tf.maximum(depth, 0)\n",
        "\n",
        "        # Prepare the mask for RGB image normalization\n",
        "        mask = tf.tile(tf.expand_dims(depth_mask, axis=-1), [1, 1, 3])\n",
        "\n",
        "        # Masked image for mean and std deviation calculation\n",
        "        masked_image = tf.boolean_mask(image, mask)\n",
        "        mean, variance = tf.nn.moments(masked_image, axes=[0])\n",
        "        std_dev = tf.sqrt(variance + 1e-6)  # Adding epsilon to avoid division by zero\n",
        "\n",
        "        # Apply the mask to image normalization\n",
        "        normalized_image = tf.where(\n",
        "            mask,\n",
        "            (image - mean) / std_dev,\n",
        "            image  # Preserve original pixels where mask is False\n",
        "        )\n",
        "\n",
        "        return normalized_image, depth\n",
        "\n",
        "    def _augment(self, image, depth):\n",
        "        target_height, target_width = 256, 256  # Set your desired fixed shape here\n",
        "\n",
        "        # Horizontal flip with 50% probability\n",
        "        if tf.random.uniform(()) > 0.5:\n",
        "            image = tf.image.flip_left_right(image)\n",
        "            depth = tf.image.flip_left_right(tf.expand_dims(depth, axis=-1))\n",
        "            depth = tf.squeeze(depth, axis=-1)\n",
        "        \n",
        "        # Vertical flip with 50% probability\n",
        "        if tf.random.uniform(()) > 0.5:\n",
        "            image = tf.image.flip_up_down(image)\n",
        "            depth = tf.image.flip_up_down(tf.expand_dims(depth, axis=-1))\n",
        "            depth = tf.squeeze(depth, axis=-1)\n",
        "        \n",
        "        # Random rotation\n",
        "        angle = tf.random.uniform([], -20, 20) * (3.14159265 / 180)  # convert degrees to radians\n",
        "        image = tfa.image.rotate(image, angle)\n",
        "        depth = tfa.image.rotate(tf.expand_dims(depth, axis=-1), angle)\n",
        "        depth = tf.squeeze(depth, axis=-1)\n",
        "        \n",
        "        # Random width shift\n",
        "        width_shift = tf.random.uniform([], -0.1, 0.1) * tf.cast(tf.shape(image)[1], tf.float32)\n",
        "        image = tfa.image.translate(image, [width_shift, 0])\n",
        "        depth = tfa.image.translate(tf.expand_dims(depth, axis=-1), [width_shift, 0])\n",
        "        depth = tf.squeeze(depth, axis=-1)\n",
        "        \n",
        "        # Random height shift\n",
        "        height_shift = tf.random.uniform([], -0.1, 0.1) * tf.cast(tf.shape(image)[0], tf.float32)\n",
        "        image = tfa.image.translate(image, [0, height_shift])\n",
        "        depth = tfa.image.translate(tf.expand_dims(depth, axis=-1), [0, height_shift])\n",
        "        depth = tf.squeeze(depth, axis=-1)\n",
        "        \n",
        "        # Random zoom\n",
        "        zoom_factor = tf.random.uniform([], 1 - 0.2, 1 + 0.2)\n",
        "        new_size = tf.cast(tf.cast(tf.shape(image)[:2], tf.float32) * zoom_factor, tf.int32)\n",
        "        image = tf.image.resize(image, new_size)\n",
        "        depth = tf.image.resize(tf.expand_dims(depth, axis=-1), new_size)\n",
        "        depth = tf.squeeze(depth, axis=-1)\n",
        "        \n",
        "        # Resize back to fixed target size\n",
        "        image = tf.image.resize(image, [target_height, target_width])\n",
        "        depth = tf.image.resize(tf.expand_dims(depth, axis=-1), [target_height, target_width])\n",
        "        depth = tf.squeeze(depth, axis=-1)\n",
        "        \n",
        "        # Random brightness adjustment\n",
        "        image = tf.image.random_brightness(image, max_delta=0.1)\n",
        "        \n",
        "        return image, depth\n",
        "\n",
        "    def load_dataset(self):\n",
        "        dataset = tf.data.TFRecordDataset(self.tfrecord_file)\n",
        "        dataset = dataset.map(self._parse_function, num_parallel_calls=tf.data.AUTOTUNE)\n",
        "        dataset = dataset.map(self._normalize, num_parallel_calls=tf.data.AUTOTUNE)\n",
        "\n",
        "        if self.augment:\n",
        "            dataset = dataset.map(self._augment, num_parallel_calls=tf.data.AUTOTUNE)\n",
        "\n",
        "        if self.shuffle:\n",
        "            dataset = dataset.shuffle(buffer_size=1000)\n",
        "\n",
        "        dataset = dataset.batch(self.batch_size)\n",
        "        dataset = dataset.prefetch(tf.data.AUTOTUNE)\n",
        "        return dataset\n",
        "    \n",
        "    def get_num_samples(self):\n",
        "        dataset = tf.data.TFRecordDataset(self.tfrecord_file)\n",
        "        count = 0\n",
        "        for _ in dataset:\n",
        "            count += 1\n",
        "        return count\n",
        "\n",
        "\n",
        "# %%\n",
        "\n",
        "class LearningRateLogger(tf.keras.callbacks.Callback):\n",
        "    def on_epoch_end(self, epoch, logs=None):\n",
        "        lr = self.model.optimizer.lr\n",
        "        if isinstance(lr, tf.keras.optimizers.schedules.LearningRateSchedule):\n",
        "            lr = lr(self.model.optimizer.iterations)\n",
        "        print(f\"Epoch {epoch + 1}: Learning rate is {tf.keras.backend.get_value(lr):.6f}\")\n",
        "\n",
        "class TensorBoardLearningRateLogger(tf.keras.callbacks.Callback):\n",
        "    def __init__(self, log_dir='./logs'):\n",
        "        super(TensorBoardLearningRateLogger, self).__init__()\n",
        "        self.file_writer = tf.summary.create_file_writer(log_dir)\n",
        "\n",
        "    def on_epoch_end(self, epoch, logs=None):\n",
        "        lr = self.model.optimizer.lr\n",
        "        if isinstance(lr, tf.keras.optimizers.schedules.LearningRateSchedule):\n",
        "            lr = lr(self.model.optimizer.iterations)\n",
        "        with self.file_writer.as_default():\n",
        "            tf.summary.scalar('learning_rate', tf.keras.backend.get_value(lr), step=epoch)\n",
        "\n",
        "# %%\n",
        "def visualize_hist(history, show=True, filename=None, title='Training history'):\n",
        "    train_hist = history.history\n",
        "\n",
        "    fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 4))\n",
        "    fig.suptitle(title, fontsize=14, fontweight='bold')\n",
        "\n",
        "    ax1.plot(train_hist['loss'])\n",
        "    ax1.plot(train_hist['val_loss'])\n",
        "    ax1.set(xlabel='epoch', ylabel='Loss')\n",
        "    ax1.legend(['train', 'valid'], loc='upper right')\n",
        "\n",
        "    ax2.plot(train_hist['mae'])\n",
        "    ax2.plot(train_hist['val_mae'])\n",
        "    ax2.set(xlabel='epoch', ylabel='MAE')\n",
        "    ax2.legend(['train', 'valid'], loc='upper right')\n",
        "\n",
        "    if show:\n",
        "        plt.show()\n",
        "\n",
        "    if filename is not None:\n",
        "        fig.savefig(filename)\n",
        "\n",
        "\n",
        "# %%\n",
        "class WarmUpCosineDecayScheduler(tf.keras.optimizers.schedules.LearningRateSchedule):\n",
        "    def __init__(self, initial_learning_rate, target_learning_rate, total_steps, warmup_steps):\n",
        "        super(WarmUpCosineDecayScheduler, self).__init__()\n",
        "        self.initial_learning_rate = initial_learning_rate\n",
        "        self.target_learning_rate = target_learning_rate\n",
        "        self.total_steps = total_steps\n",
        "        self.warmup_steps = warmup_steps\n",
        "        self.decay_steps = total_steps - warmup_steps\n",
        "\n",
        "    def __call__(self, step):\n",
        "        # Convert to float32 to ensure the operations are compatible\n",
        "        step = tf.cast(step, tf.float32)\n",
        "        warmup_steps = tf.cast(self.warmup_steps, tf.float32)\n",
        "        decay_steps = tf.cast(self.decay_steps, tf.float32)\n",
        "\n",
        "        # Compute the warmup learning rate\n",
        "        warmup_lr = self.initial_learning_rate + (self.target_learning_rate - self.initial_learning_rate) * (step / warmup_steps)\n",
        "\n",
        "        # Compute the cosine decay learning rate\n",
        "        cosine_decay = 0.5 * (1 + tf.cos(np.pi * (step - warmup_steps) / decay_steps))\n",
        "        decayed_lr = (self.target_learning_rate - self.initial_learning_rate) * cosine_decay + self.initial_learning_rate\n",
        "\n",
        "        # Choose the learning rate based on the step\n",
        "        learning_rate = tf.where(step < warmup_steps, warmup_lr, decayed_lr)\n",
        "        return learning_rate\n",
        "\n",
        "    def get_config(self):\n",
        "        return {\n",
        "            'initial_learning_rate': self.initial_learning_rate,\n",
        "            'target_learning_rate': self.target_learning_rate,\n",
        "            'total_steps': self.total_steps,\n",
        "            'warmup_steps': self.warmup_steps\n",
        "        }\n",
        "\n",
        "# %%\n",
        "# Parameters\n",
        "initial_learning_rate = 1e-5\n",
        "target_learning_rate = 1e-2\n",
        "verbose = 1\n",
        "shuffle = True\n",
        "checkpoint = './best_model.h5'\n",
        "\n",
        "trainloader = TFRecordDataHandler('train.tfrecords', batch_size=batch_size, shuffle=shuffle, augment=True)\n",
        "valloader = TFRecordDataHandler('validation.tfrecords', batch_size=batch_size, shuffle=shuffle, augment=False)\n",
        "train_dg = trainloader.load_dataset()\n",
        "validation_dg = valloader.load_dataset()\n",
        "\n",
        "num_samples_train = trainloader.get_num_samples()\n",
        "\n",
        "print('Number of data used to train:')\n",
        "print(num_samples_train * batch_size)\n",
        "# print('')\n",
        "\n",
        "for x, y in train_dg.take(1):\n",
        "    print(\"Input shape:\", x.shape)  # Should be (batch_size, height, width, channels)\n",
        "    print(\"Label shape:\", y.shape)  # Should be (batch_size, ...) depending on your task\n",
        "\n",
        "\n",
        "total_steps = num_samples_train // num_samples_train * n_epochs\n",
        "\n",
        "\n",
        "import datetime\n",
        "for wu_ratio in [0.2]:\n",
        "\n",
        "    warmup_steps = int(wu_ratio * total_steps)\n",
        "\n",
        "    folder_name = f'transunet_DA'\n",
        "     #get the time of the experiment\n",
        "    time = datetime.datetime.now().strftime(\"%Y%m%d-%H%M%S\")\n",
        "    experiment_name = folder_name + '_' + time\n",
        "    tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=os.path.join(\"logs\",  experiment_name))\n",
        "    tensorboard_lr_logger = TensorBoardLearningRateLogger(os.path.join(\"logs\", experiment_name))\n",
        "\n",
        "\n",
        "    # Create the learning rate schedule\n",
        "    lr_schedule = WarmUpCosineDecayScheduler(\n",
        "        initial_learning_rate=initial_learning_rate,\n",
        "        target_learning_rate=target_learning_rate,\n",
        "        total_steps=total_steps,\n",
        "        warmup_steps=warmup_steps\n",
        "    )\n",
        "\n",
        "    # Define the optimizer with the custom learning rate schedule\n",
        "    optimizer = tf.keras.optimizers.Adam(learning_rate=lr_schedule)\n",
        "\n",
        "    model = models.transunet_2d((256, 256, 3), [64, 128, 256, 512, 1024], n_labels=1,\n",
        "                                stack_num_down=2, stack_num_up=1, \n",
        "                                activation='GELU', output_activation='Sigmoid',\n",
        "                                batch_norm=True, pool='max', unpool=False, name='transunet_2d', embed_dim=100, num_heads=4)\n",
        "\n",
        "    print('Learning rate:', initial_learning_rate)\n",
        "    # defining the optimizer\n",
        "    model.compile(optimizer, loss=tf.keras.losses.MeanSquaredError(), metrics=['mae'])\n",
        "\n",
        "    es = tf.keras.callbacks.EarlyStopping(monitor='val_mae', patience=3)\n",
        "\n",
        "    # saving the best model based on val_loss\n",
        "    mc = tf.keras.callbacks.ModelCheckpoint(os.path.join(\"models\", experiment_name, checkpoint), monitor='val_mae', mode='min', save_best_only=True)\n",
        "    \n",
        "    # training the model and saving the history (.take(10))\n",
        "    history = model.fit(train_dg, validation_data=validation_dg, epochs=n_epochs, verbose=verbose, workers=-1, use_multiprocessing=True,\n",
        "                        callbacks=[tensorboard_callback, tensorboard_lr_logger, mc, es])\n",
        "\n",
        "    # Plot and write the history\n",
        "    if not os.path.exists(experiment_name):\n",
        "        os.makedirs(experiment_name)\n",
        "    filename = os.path.join(experiment_name, 'train_history.jpg')\n",
        "    visualize_hist(history, show=True, filename=filename)\n",
        "\n",
        "    #with open('train_history_lr_'+str(lr)+'.pkl', 'wb') as handle:\n",
        "    #    pickle.dump(history.history, handle, protocol=pickle.HIGHEST_PROTOCOL)\n",
        "    # Get the best validation loss and the epoch\n",
        "\n",
        "    best_val_loss = min(history.history['val_loss'])\n",
        "    best_epoch = history.history['val_loss'].index(best_val_loss)\n",
        "    # d_lr[lr] = (best_val_loss, best_epoch)\n",
        "    print('Best validation loss:', best_val_loss, 'at epoch', best_epoch)\n",
        "\n",
        "    # Clear gpu memory\n",
        "    del model\n",
        "    gc.collect()\n",
        "    tf.keras.backend.clear_session()\n",
        "\n",
        "# %%"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": []
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "gpuType": "T4",
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "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": "3.11.5"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}