{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.12","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"gpu","dataSources":[{"sourceType":"competition","sourceId":4104,"databundleVersionId":46661},{"sourceType":"datasetVersion","sourceId":952401,"datasetId":517172,"databundleVersionId":980293}],"dockerImageVersionId":30588,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"from tensorflow import lite\nimport tensorflow as tf\nfrom tensorflow import keras\nfrom tensorflow.keras import layers\nimport numpy as np\nimport pandas as pd\nimport random, os\nimport shutil\nimport matplotlib.pyplot as plt\nfrom matplotlib.image import imread\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom tensorflow.keras.metrics import categorical_accuracy\nfrom sklearn.model_selection import train_test_split","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:48:59.173343Z","iopub.execute_input":"2026-05-05T18:48:59.174034Z","iopub.status.idle":"2026-05-05T18:48:59.178867Z","shell.execute_reply.started":"2026-05-05T18:48:59.174003Z","shell.execute_reply":"2026-05-05T18:48:59.177920Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"df = pd.read_csv(r'../input/diabetic-retinopathy-224x224-gaussian-filtered/train.csv')\n\ndiagnosis_dict_binary = {\n    0: 'No_DR',\n    1: 'DR',\n    2: 'DR',\n    3: 'DR',\n    4: 'DR'\n}\n\ndiagnosis_dict = {\n    0: 'No_DR',\n    1: 'Mild',\n    2: 'Moderate',\n    3: 'Severe',\n    4: 'Proliferate_DR',\n}\n\n\ndf['binary_type'] =  df['diagnosis'].map(diagnosis_dict_binary.get)\ndf['type'] = df['diagnosis'].map(diagnosis_dict.get)\ndf.head()","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:49:07.039098Z","iopub.execute_input":"2026-05-05T18:49:07.039724Z","iopub.status.idle":"2026-05-05T18:49:07.074352Z","shell.execute_reply.started":"2026-05-05T18:49:07.039693Z","shell.execute_reply":"2026-05-05T18:49:07.073584Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"df['type'].value_counts().plot(kind='barh')","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:49:13.272947Z","iopub.execute_input":"2026-05-05T18:49:13.273712Z","iopub.status.idle":"2026-05-05T18:49:13.522822Z","shell.execute_reply.started":"2026-05-05T18:49:13.273683Z","shell.execute_reply":"2026-05-05T18:49:13.521934Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_intermediate, val = train_test_split(df, test_size = 0.15, stratify = df['type'])\ntrain, test = train_test_split(train_intermediate, test_size = 0.15 / (1 - 0.15), stratify = train_intermediate['type'])\n\nprint(\"For Training Dataset :\")\nprint(train['type'].value_counts(), '\\n')\nprint(\"For Testing Dataset :\")\nprint(test['type'].value_counts(), '\\n')\nprint(\"For Validation Dataset :\")\nprint(val['type'].value_counts(), '\\n')","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:49:18.663478Z","iopub.execute_input":"2026-05-05T18:49:18.663950Z","iopub.status.idle":"2026-05-05T18:49:18.687823Z","shell.execute_reply.started":"2026-05-05T18:49:18.663910Z","shell.execute_reply":"2026-05-05T18:49:18.686946Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"base_dir = ''\n\ntrain_dir = os.path.join(base_dir, 'train')\nval_dir = os.path.join(base_dir, 'val')\ntest_dir = os.path.join(base_dir, 'test')\n\nif os.path.exists(base_dir):\n    shutil.rmtree(base_dir)\n\nif os.path.exists(train_dir):\n    shutil.rmtree(train_dir)\nos.makedirs(train_dir)\n\nif os.path.exists(val_dir):\n    shutil.rmtree(val_dir)\nos.makedirs(val_dir)\n\nif os.path.exists(test_dir):\n    shutil.rmtree(test_dir)\nos.makedirs(test_dir)","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:49:24.869399Z","iopub.execute_input":"2026-05-05T18:49:24.870022Z","iopub.status.idle":"2026-05-05T18:49:24.876224Z","shell.execute_reply.started":"2026-05-05T18:49:24.869994Z","shell.execute_reply":"2026-05-05T18:49:24.875357Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"src_dir = r'../input/diabetic-retinopathy-224x224-gaussian-filtered/gaussian_filtered_images/gaussian_filtered_images'\nfor index, row in train.iterrows():\n    diagnosis = row['type']\n    binary_diagnosis = row['binary_type']\n    id_code = row['id_code'] + \".png\"\n    srcfile = os.path.join(src_dir, diagnosis, id_code)\n    dstfile = os.path.join(train_dir, binary_diagnosis)\n    os.makedirs(dstfile, exist_ok = True)\n    shutil.copy(srcfile, dstfile)\n\nfor index, row in val.iterrows():\n    diagnosis = row['type']\n    binary_diagnosis = row['binary_type']\n    id_code = row['id_code'] + \".png\"\n    srcfile = os.path.join(src_dir, diagnosis, id_code)\n    dstfile = os.path.join(val_dir, binary_diagnosis)\n    os.makedirs(dstfile, exist_ok = True)\n    shutil.copy(srcfile, dstfile)\n \nfor index, row in test.iterrows():\n    diagnosis = row['type']\n    binary_diagnosis = row['binary_type']\n    id_code = row['id_code'] + \".png\"\n    srcfile = os.path.join(src_dir, diagnosis, id_code)\n    dstfile = os.path.join(test_dir, binary_diagnosis)\n    os.makedirs(dstfile, exist_ok = True)\n    shutil.copy(srcfile, dstfile)","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:49:29.213414Z","iopub.execute_input":"2026-05-05T18:49:29.214241Z","iopub.status.idle":"2026-05-05T18:49:54.271407Z","shell.execute_reply.started":"2026-05-05T18:49:29.214210Z","shell.execute_reply":"2026-05-05T18:49:54.270712Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_path = 'train'\nval_path = 'val'\ntest_path = 'test'\n\ntrain_batches = ImageDataGenerator(rescale = 1./255).flow_from_directory(train_path, target_size=(224,224), shuffle = True)\nval_batches = ImageDataGenerator(rescale = 1./255).flow_from_directory(val_path, target_size=(224,224), shuffle = True)\ntest_batches = ImageDataGenerator(rescale = 1./255).flow_from_directory(test_path, target_size=(224,224), shuffle = False)","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:50:02.084641Z","iopub.execute_input":"2026-05-05T18:50:02.084952Z","iopub.status.idle":"2026-05-05T18:50:02.204757Z","shell.execute_reply.started":"2026-05-05T18:50:02.084931Z","shell.execute_reply":"2026-05-05T18:50:02.203718Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model = tf.keras.Sequential([\n    layers.Conv2D(8, (3,3), padding=\"valid\", input_shape=(224,224,3), activation = 'relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    \n    layers.Conv2D(16, (3,3), padding=\"valid\", activation = 'relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    \n    layers.Conv2D(32, (4,4), padding=\"valid\", activation = 'relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    \n    layers.Conv2D(64, (4,4), padding=\"valid\", activation = 'relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n \n    layers.Flatten(),\n    layers.Dense(64, activation = 'relu'),\n    layers.Dropout(0.15),\n    layers.Dense(2, activation = 'softmax')\n])\n\nmodel.compile(optimizer=tf.keras.optimizers.Adam(lr = 1e-5),\n              loss=tf.keras.losses.BinaryCrossentropy(),\n              metrics=['acc'])\n\nhistory = model.fit(train_batches,\n                    epochs=15,\n                    validation_data=val_batches)","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:50:08.095068Z","iopub.execute_input":"2026-05-05T18:50:08.096057Z","iopub.status.idle":"2026-05-05T18:52:04.791265Z","shell.execute_reply.started":"2026-05-05T18:50:08.096027Z","shell.execute_reply":"2026-05-05T18:52:04.790459Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# Standard Dropout","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model = tf.keras.Sequential([\n    layers.Conv2D(8, (3,3), padding=\"valid\", input_shape=(224,224,3), activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.Dropout(0.25),\n\n    layers.Conv2D(16, (3,3), padding=\"valid\", activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.Dropout(0.25),\n\n    layers.Conv2D(32, (4,4), padding=\"valid\", activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.Dropout(0.3),\n\n    layers.Conv2D(64, (4,4), padding=\"valid\", activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.Dropout(0.3),\n\n    layers.Flatten(),\n    layers.Dense(64, activation='relu'),\n    layers.Dropout(0.5),   # standard for dense layer\n    layers.Dense(2, activation='softmax')\n])\n\nmodel.compile(\n    optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5),\n    loss=tf.keras.losses.BinaryCrossentropy(),\n    metrics=['accuracy']\n)\n\nhistory = model.fit(\n    train_batches,\n    epochs=15,\n    validation_data=val_batches\n)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-05-05T18:52:18.993858Z","iopub.execute_input":"2026-05-05T18:52:18.994165Z","iopub.status.idle":"2026-05-05T18:54:12.332749Z","shell.execute_reply.started":"2026-05-05T18:52:18.994143Z","shell.execute_reply":"2026-05-05T18:54:12.331990Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# Spatial Dropout","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model = tf.keras.Sequential([\n    layers.Conv2D(8, (3,3), padding=\"valid\", input_shape=(224,224,3), activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.SpatialDropout2D(0.2),\n\n    layers.Conv2D(16, (3,3), padding=\"valid\", activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.SpatialDropout2D(0.2),\n\n    layers.Conv2D(32, (4,4), padding=\"valid\", activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.SpatialDropout2D(0.3),\n\n    layers.Conv2D(64, (4,4), padding=\"valid\", activation='relu'),\n    layers.MaxPooling2D(pool_size=(2,2)),\n    layers.BatchNormalization(),\n    layers.SpatialDropout2D(0.3),\n\n    layers.Flatten(),\n    layers.Dense(64, activation='relu'),\n    layers.Dropout(0.5),   # keep standard dropout for dense layer\n    layers.Dense(2, activation='softmax')\n])\n\nmodel.compile(\n    optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5),\n    loss=tf.keras.losses.BinaryCrossentropy(),\n    metrics=['accuracy']\n)\n\nhistory = model.fit(\n    train_batches,\n    epochs=15,\n    validation_data=val_batches\n)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-05-05T18:54:48.502909Z","iopub.execute_input":"2026-05-05T18:54:48.503785Z","iopub.status.idle":"2026-05-05T18:56:46.342828Z","shell.execute_reply.started":"2026-05-05T18:54:48.503754Z","shell.execute_reply":"2026-05-05T18:56:46.342073Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model.save('64x3-CNN.model')","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:57:59.304382Z","iopub.execute_input":"2026-05-05T18:57:59.305147Z","iopub.status.idle":"2026-05-05T18:58:01.615746Z","shell.execute_reply.started":"2026-05-05T18:57:59.305115Z","shell.execute_reply":"2026-05-05T18:58:01.614742Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"loss, acc = model.evaluate_generator(test_batches, verbose=1)\nprint(\"Loss: \", loss)\nprint(\"Accuracy: \", acc)","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:58:04.928911Z","iopub.execute_input":"2026-05-05T18:58:04.929917Z","iopub.status.idle":"2026-05-05T18:58:06.526286Z","shell.execute_reply.started":"2026-05-05T18:58:04.929873Z","shell.execute_reply":"2026-05-05T18:58:06.525184Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import tensorflow as tf\nimport cv2\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n\ndef predict_class(path):\n    img = cv2.imread(path)\n\n    RGBImg = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)\n    RGBImg= cv2.resize(RGBImg,(224,224))\n    plt.imshow(RGBImg)\n    image = np.array(RGBImg) / 255.0\n    new_model = tf.keras.models.load_model(\"64x3-CNN.model\")\n    predict=new_model.predict(np.array([image]))\n    per=np.argmax(predict,axis=1)\n    if per==1:\n        print('Diabetic Retinopathy Not Detected')\n    else:\n        print('Diabetic Retinopathy Detected')","metadata":{"execution":{"iopub.status.busy":"2026-05-05T18:58:17.189363Z","iopub.execute_input":"2026-05-05T18:58:17.190222Z","iopub.status.idle":"2026-05-05T18:58:17.421982Z","shell.execute_reply.started":"2026-05-05T18:58:17.190192Z","shell.execute_reply":"2026-05-05T18:58:17.420708Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"predict_class('/kaggle/input/diabetic-retinopathy-224x224-gaussian-filtered/gaussian_filtered_images/gaussian_filtered_images/Severe/1b495ac025b7.png')","metadata":{"execution":{"iopub.status.busy":"2026-05-05T16:03:11.444004Z","iopub.execute_input":"2026-05-05T16:03:11.444672Z","iopub.status.idle":"2026-05-05T16:03:12.733952Z","shell.execute_reply.started":"2026-05-05T16:03:11.444643Z","shell.execute_reply":"2026-05-05T16:03:12.733085Z"},"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}