3. Data_Augmentation.ipynb

{"cells":[{"cell_type":"markdown","metadata":{"id":"-YMDkQPLzRRo"},"source":["Import Libraries"]},{"cell_type":"code","execution_count":1,"metadata":{"id":"DKO85oCYqjdK"},"outputs":[],"source":["import tensorflow as tf\n","import tensorflow.keras\n","import matplotlib.pylab as plt\n","import numpy as np\n","import os\n","import shutil\n","import albumentations as A\n","import cv2\n","from sklearn.model_selection import train_test_split\n","\n","SEED = 42\n","import random \n","random.seed(SEED)\n","import numpy as np\n","np.random.seed(SEED)\n","import tensorflow as tf\n","tf.random.set_seed(SEED)"]},{"cell_type":"markdown","metadata":{},"source":["# ** Inputs ** "]},{"cell_type":"code","execution_count":2,"metadata":{},"outputs":[],"source":["# New Augmented Folder name\n","new_folder_name = 'FMD_aug_tvt'\n","\n","# Name of folder where the regular folders are\n","source_folder = 'FMD_tvt'\n","\n","# Mulitplier for the train images\n","train_multiplier = 10\n","\n","# Mulitplier for the train images\n","val_multiplier = 5"]},{"cell_type":"code","execution_count":3,"metadata":{},"outputs":[{"name":"stdout","output_type":"stream","text":["['fabric', 'foliage', 'glass', 'leather', 'metal', 'paper', 'plastic', 'stone', 'water', 'wood']\n"]}],"source":["tvt_folders = ['train', 'val', 'test']\n","im_folders = ['image', 'mask']\n","\n","Fmd_class_path = os.path.join(source_folder, tvt_folders[0], im_folders[0])\n","classes = os.listdir(os.path.join(Fmd_class_path))\n","print(classes)"]},{"cell_type":"markdown","metadata":{"id":"QX5AkhpSAaMd"},"source":["# Augment Function"]},{"cell_type":"code","execution_count":4,"metadata":{"id":"-Z9D8zPSABAl"},"outputs":[],"source":["transform = A.Compose([\n","    A.RandomCrop(width=256, height=256),\n","    A.HorizontalFlip(p=0.5),\n","    A.VerticalFlip(p=0.5),\n","    A.RGBShift(p=0.4),\n","    A.ColorJitter(p=0.2),\n","    A.RandomBrightnessContrast(p=0.2)\n","])"]},{"cell_type":"markdown","metadata":{"id":"ddfvCmyvAxkD"},"source":["# Making the new folder"]},{"cell_type":"code","execution_count":5,"metadata":{"id":"DOeGxQUSAwgM"},"outputs":[],"source":["def make_FMD_tvt(file_name, classes, tvt_folders, im_folders):\n","    if not os.path.exists(os.path.join(file_name)):\n","        for tvt in tvt_folders:\n","            for im in im_folders:\n","                for class_ in classes:\n","                    os.makedirs(os.path.join(file_name, tvt, im, class_))"]},{"cell_type":"markdown","metadata":{},"source":["Running the function"]},{"cell_type":"code","execution_count":6,"metadata":{},"outputs":[],"source":["make_FMD_tvt(new_folder_name, classes, tvt_folders, im_folders)"]},{"cell_type":"markdown","metadata":{"id":"VfkEQIu0AdD3"},"source":["# Splitting Train and Test from FMD to New folder"]},{"cell_type":"code","execution_count":7,"metadata":{"id":"mIqeQkui_Ip-"},"outputs":[],"source":["def move_images_to_new(class_name, new_folder_name, tvt_folders, im_folders):\n","    train_imgs = os.listdir(os.path.join(source_folder, tvt_folders[0], im_folders[0], class_name))\n","    val_imgs = os.listdir(os.path.join(source_folder, tvt_folders[1], im_folders[0], class_name))\n","    test_imgs = os.listdir(os.path.join(source_folder, tvt_folders[2], im_folders[0], class_name))\n","\n","    tvt_list = [train_imgs, val_imgs, test_imgs]\n","\n","    for i in range(len(tvt_folders)):\n","        for j in range(len(tvt_list[i])):\n","            source_path = os.path.join(source_folder, tvt_folders[i], im_folders[0], class_name, tvt_list[i][j])\n","            destination_path = os.path.join(new_folder_name, tvt_folders[i], im_folders[0], class_name)\n","            shutil.copy(source_path, destination_path)\n","    return\n","  "]},{"cell_type":"markdown","metadata":{},"source":["Running function"]},{"cell_type":"code","execution_count":8,"metadata":{},"outputs":[],"source":["#Split train stuff accordingly\n","for i in range(len(classes)):\n","  move_images_to_new(classes[i], new_folder_name, tvt_folders, im_folders)"]},{"cell_type":"markdown","metadata":{},"source":["# Moving the masks to New Folder"]},{"cell_type":"code","execution_count":9,"metadata":{"id":"LAmWIH_qkhYj"},"outputs":[],"source":["def move_masks(class_name, new_folder_name, tvt_folders, im_folders):\n","    train_list = os.listdir(os.path.join(new_folder_name, tvt_folders[0], im_folders[0], class_name))\n","    val_list = os.listdir(os.path.join(new_folder_name, tvt_folders[1], im_folders[0], class_name))\n","    test_list = os.listdir(os.path.join(new_folder_name, tvt_folders[2], im_folders[0], class_name))\n","\n","    tvt_list = [train_list, val_list, test_list]\n","  \n","    for i in range(len(tvt_list)):\n","        for j in range(len(tvt_list[i])):\n","            source_path = os.path.join(source_folder, tvt_folders[i], im_folders[1], class_name, tvt_list[i][j])\n","            destination_path = os.path.join(new_folder_name, tvt_folders[i], im_folders[1], class_name)\n","            if os.path.exists(source_path):\n","                shutil.copy(source_path, destination_path)\n","    return"]},{"cell_type":"markdown","metadata":{},"source":["Running Function"]},{"cell_type":"code","execution_count":10,"metadata":{},"outputs":[],"source":["#Split train stuff accordingly\n","for i in range(len(classes)):\n","  move_masks(classes[i], new_folder_name, tvt_folders, im_folders)"]},{"cell_type":"markdown","metadata":{},"source":["# Data Augmentation Function"]},{"cell_type":"code","execution_count":11,"metadata":{"executionInfo":{"elapsed":403,"status":"ok","timestamp":1667022662856,"user":{"displayName":"Maxime Capelle","userId":"11180759485503328219"},"user_tz":-480},"id":"mNVlpLxDqxxj"},"outputs":[],"source":["def data_augmentation(class_name, new_folder_name, tvt_folders, im_folders, multiplier, tvt_num):\n","    print(f'Starting {class_name}')\n","    images_in_folder = os.listdir(os.path.join(new_folder_name, tvt_folders[tvt_num], im_folders[0], class_name))\n","    for image_name in images_in_folder:\n","        img = cv2.imread(os.path.join(new_folder_name, tvt_folders[tvt_num], im_folders[0], class_name, image_name))\n","        msk = cv2.imread(os.path.join(new_folder_name, tvt_folders[tvt_num], im_folders[1], class_name, image_name))\n","\n","        store_path_image = os.path.join(new_folder_name, tvt_folders[tvt_num], im_folders[0], class_name)\n","        store_path_mask = os.path.join(new_folder_name, tvt_folders[tvt_num], im_folders[1], class_name)\n","\n","        for i in range(multiplier):\n","            transformed = transform(image = img, mask = msk)\n","      \n","            transformed_image = transformed['image']\n","            transformed_mask = transformed['mask']\n","\n","            cv2.imwrite(os.path.join(store_path_image, f'{image_name[0:-4]}_{i}.jpg'), transformed_image)\n","            cv2.imwrite(os.path.join(store_path_mask, f'{image_name[0:-4]}_{i}.jpg'), transformed_mask)\n","    \n","    print(f\"Success {class_name}\")\n","        "]},{"cell_type":"markdown","metadata":{"id":"nBR0BIxjdgBa"},"source":["Running Function"]},{"cell_type":"code","execution_count":12,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":3006,"status":"ok","timestamp":1667022724437,"user":{"displayName":"Maxime Capelle","userId":"11180759485503328219"},"user_tz":-480},"id":"n2E1NG_rxAFN","outputId":"d0429286-f907-4c1c-85ae-f3d5e700621d"},"outputs":[{"name":"stdout","output_type":"stream","text":["Starting fabric\n","Success fabric\n","Starting fabric\n","Success fabric\n","Starting foliage\n","Success foliage\n","Starting foliage\n","Success foliage\n","Starting glass\n","Success glass\n","Starting glass\n","Success glass\n","Starting leather\n","Success leather\n","Starting leather\n","Success leather\n","Starting metal\n","Success metal\n","Starting metal\n","Success metal\n","Starting paper\n","Success paper\n","Starting paper\n","Success paper\n","Starting plastic\n","Success plastic\n","Starting plastic\n","Success plastic\n","Starting stone\n","Success stone\n","Starting stone\n","Success stone\n","Starting water\n","Success water\n","Starting water\n","Success water\n","Starting wood\n","Success wood\n","Starting wood\n","Success wood\n"]}],"source":["for i in range(len(classes)):\n","  data_augmentation(classes[i], new_folder_name, tvt_folders, im_folders, train_multiplier, 0)\n","  data_augmentation(classes[i], new_folder_name, tvt_folders, im_folders, val_multiplier, 1)"]}],"metadata":{"colab":{"collapsed_sections":[],"provenance":[]},"kernelspec":{"display_name":"Python 3.10.4","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":"3.10.4"},"vscode":{"interpreter":{"hash":"b942636a785217c0b780635fcfa4dacafb5d46c802d84d4ffbd090c16749e713"}}},"nbformat":4,"nbformat_minor":0}