UNET-Image-Segmentation-PASCAL-VOC.ipynb

{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.7.10","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import numpy as np\nimport tensorflow as tf\nimport matplotlib.pyplot as plt\nfrom PIL import Image\nimport os\nfrom IPython.display import clear_output\n","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2021-10-02T11:55:30.039544Z","iopub.execute_input":"2021-10-02T11:55:30.039931Z","iopub.status.idle":"2021-10-02T11:55:34.234824Z","shell.execute_reply.started":"2021-10-02T11:55:30.039833Z","shell.execute_reply":"2021-10-02T11:55:34.234076Z"},"trusted":true},"execution_count":1,"outputs":[]},{"cell_type":"code","source":"IMG_HEIGHT = 128\nIMG_WIDTH  = 128","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:55:34.236453Z","iopub.execute_input":"2021-10-02T11:55:34.236710Z","iopub.status.idle":"2021-10-02T11:55:34.244742Z","shell.execute_reply.started":"2021-10-02T11:55:34.236677Z","shell.execute_reply":"2021-10-02T11:55:34.244019Z"},"trusted":true},"execution_count":2,"outputs":[]},{"cell_type":"code","source":"def Read_All_Data(path):\n    temp = []\n    updated_path = os.path.join(path,\"VOC2012\",\"ImageSets\",\"Segmentation\",\"trainval.txt\")\n    with open(updated_path,\"r\") as file_:\n        Instances = file_.read().split()\n        img_arr = []\n        seg_arr = []\n        for img in Instances:\n            path_img = os.path.join(path,\"VOC2012\",\"JPEGImages\",img+\".jpg\")\n            path_seg = os.path.join(path,\"VOC2012\",\"SegmentationClass\",img+\".png\")\n            img = np.array(Image.open(path_img).resize((IMG_WIDTH, IMG_HEIGHT)))\n            seg = np.array(Image.open(path_seg).resize((IMG_WIDTH, IMG_HEIGHT)))\n            seg[seg == 255] = 21\n            img_arr.append(img)\n            seg_arr.append(seg)\n    return np.array(img_arr), np.array(seg_arr)","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:55:34.246166Z","iopub.execute_input":"2021-10-02T11:55:34.246432Z","iopub.status.idle":"2021-10-02T11:55:34.255659Z","shell.execute_reply.started":"2021-10-02T11:55:34.246395Z","shell.execute_reply":"2021-10-02T11:55:34.254835Z"},"trusted":true},"execution_count":3,"outputs":[]},{"cell_type":"code","source":"path = \"../input/pascal-voc-2012\"\nfull_dataset = tf.data.Dataset.from_tensor_slices(Read_All_Data(path))","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:55:34.257296Z","iopub.execute_input":"2021-10-02T11:55:34.257592Z","iopub.status.idle":"2021-10-02T11:56:41.172144Z","shell.execute_reply.started":"2021-10-02T11:55:34.257556Z","shell.execute_reply":"2021-10-02T11:56:41.170148Z"},"trusted":true},"execution_count":4,"outputs":[]},{"cell_type":"code","source":"DATASET_SIZE = 2913","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.174731Z","iopub.execute_input":"2021-10-02T11:56:41.175042Z","iopub.status.idle":"2021-10-02T11:56:41.179244Z","shell.execute_reply.started":"2021-10-02T11:56:41.174996Z","shell.execute_reply":"2021-10-02T11:56:41.178282Z"},"trusted":true},"execution_count":5,"outputs":[]},{"cell_type":"code","source":"train_size = int(0.7 * DATASET_SIZE)\nval_size = int(0.2 * DATASET_SIZE)\ntest_size = int(0.1 * DATASET_SIZE)\n\nfull_dataset = full_dataset.shuffle(buffer_size = DATASET_SIZE)\ntrain_dataset = full_dataset.take(train_size)\ntest_dataset = full_dataset.skip(train_size)\nval_dataset = test_dataset.skip(test_size)\ntest_dataset = test_dataset.take(test_size)","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.181211Z","iopub.execute_input":"2021-10-02T11:56:41.181488Z","iopub.status.idle":"2021-10-02T11:56:41.197926Z","shell.execute_reply.started":"2021-10-02T11:56:41.181449Z","shell.execute_reply":"2021-10-02T11:56:41.196968Z"},"trusted":true},"execution_count":6,"outputs":[]},{"cell_type":"code","source":"def normalize(input_image, input_mask):\n    input_image = tf.cast(input_image, tf.float32) / 255.0\n    return input_image, input_mask","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.199454Z","iopub.execute_input":"2021-10-02T11:56:41.199755Z","iopub.status.idle":"2021-10-02T11:56:41.204539Z","shell.execute_reply.started":"2021-10-02T11:56:41.199705Z","shell.execute_reply":"2021-10-02T11:56:41.203544Z"},"trusted":true},"execution_count":7,"outputs":[]},{"cell_type":"code","source":"def load_image_train(input_image, input_mask):\n    input_mask = input_mask[ ..., np.newaxis]\n    \n\n    # random flip the image and mask\n    if tf.random.uniform(()) > 0.5:\n        input_image = tf.image.flip_left_right(input_image)\n        input_mask = tf.image.flip_left_right(input_mask)\n    \n    input_mask = input_mask[:, :, 0]\n    input_image, input_mask = normalize(input_image, input_mask)\n\n    return input_image, input_mask","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.206394Z","iopub.execute_input":"2021-10-02T11:56:41.206711Z","iopub.status.idle":"2021-10-02T11:56:41.214157Z","shell.execute_reply.started":"2021-10-02T11:56:41.206648Z","shell.execute_reply":"2021-10-02T11:56:41.213240Z"},"trusted":true},"execution_count":8,"outputs":[]},{"cell_type":"code","source":"def load_image_test(input_image, input_mask):\n    input_mask = input_mask[np.newaxis, ...]\n    \n    \n    input_mask = input_mask[0]\n    input_image, input_mask = normalize(input_image, input_mask)\n\n    return input_image, input_mask","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.215808Z","iopub.execute_input":"2021-10-02T11:56:41.216212Z","iopub.status.idle":"2021-10-02T11:56:41.222662Z","shell.execute_reply.started":"2021-10-02T11:56:41.216172Z","shell.execute_reply":"2021-10-02T11:56:41.221827Z"},"trusted":true},"execution_count":9,"outputs":[]},{"cell_type":"code","source":"BATCH_SIZE = 32\nBUFFER_SIZE = 1000\nSTEPS_PER_EPOCH = train_size // BATCH_SIZE","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.224584Z","iopub.execute_input":"2021-10-02T11:56:41.224906Z","iopub.status.idle":"2021-10-02T11:56:41.234110Z","shell.execute_reply.started":"2021-10-02T11:56:41.224872Z","shell.execute_reply":"2021-10-02T11:56:41.233315Z"},"trusted":true},"execution_count":10,"outputs":[]},{"cell_type":"code","source":"train = train_dataset.map(load_image_train, num_parallel_calls=tf.data.experimental.AUTOTUNE)\nval   = train_dataset.map(load_image_test, num_parallel_calls=tf.data.experimental.AUTOTUNE)\ntest = test_dataset.map(load_image_test, num_parallel_calls=tf.data.experimental.AUTOTUNE)","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.235752Z","iopub.execute_input":"2021-10-02T11:56:41.236058Z","iopub.status.idle":"2021-10-02T11:56:41.464378Z","shell.execute_reply.started":"2021-10-02T11:56:41.236021Z","shell.execute_reply":"2021-10-02T11:56:41.463556Z"},"trusted":true},"execution_count":11,"outputs":[]},{"cell_type":"code","source":"train_dataset = train.cache().shuffle(BUFFER_SIZE).batch(BATCH_SIZE).repeat()\ntrain_dataset = train_dataset.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)\ntest_dataset = test.batch(BATCH_SIZE)\nval_dataset = val.batch(BATCH_SIZE)","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.465849Z","iopub.execute_input":"2021-10-02T11:56:41.466104Z","iopub.status.idle":"2021-10-02T11:56:41.476168Z","shell.execute_reply.started":"2021-10-02T11:56:41.466070Z","shell.execute_reply":"2021-10-02T11:56:41.475297Z"},"trusted":true},"execution_count":12,"outputs":[]},{"cell_type":"code","source":"for image, mask in train.take(2):\n    sample_image, sample_mask = image, mask\n    plt.imshow(sample_image)\n    print(sample_image.shape)\n    plt.show()\n    plt.imshow(sample_mask)\n    plt.show()","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:41.477938Z","iopub.execute_input":"2021-10-02T11:56:41.478309Z","iopub.status.idle":"2021-10-02T11:56:42.677285Z","shell.execute_reply.started":"2021-10-02T11:56:41.478273Z","shell.execute_reply":"2021-10-02T11:56:42.676573Z"},"trusted":true},"execution_count":13,"outputs":[]},{"cell_type":"code","source":"base_model = tf.keras.applications.MobileNetV2(input_shape=[128, 128, 3], include_top=False)","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:42.681363Z","iopub.execute_input":"2021-10-02T11:56:42.681597Z","iopub.status.idle":"2021-10-02T11:56:44.596573Z","shell.execute_reply.started":"2021-10-02T11:56:42.681572Z","shell.execute_reply":"2021-10-02T11:56:44.595811Z"},"trusted":true},"execution_count":14,"outputs":[]},{"cell_type":"code","source":"base_model.summary()","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:44.598059Z","iopub.execute_input":"2021-10-02T11:56:44.598323Z","iopub.status.idle":"2021-10-02T11:56:44.679130Z","shell.execute_reply.started":"2021-10-02T11:56:44.598292Z","shell.execute_reply":"2021-10-02T11:56:44.678287Z"},"trusted":true},"execution_count":15,"outputs":[]},{"cell_type":"code","source":"layer_names = [\n    'block_1_expand_relu',   # 64x64\n    'block_3_expand_relu',   # 32x32\n    'block_6_expand_relu',   # 16x16\n    'block_13_expand_relu',  # 8x8\n    'block_16_project',      # 4x4\n]\nbase_model_outputs = [base_model.get_layer(name).output for name in layer_names ]\ndown_stack = tf.keras.Model(inputs=base_model.input, outputs=base_model_outputs)\n\ndown_stack.trainable = False","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:44.681250Z","iopub.execute_input":"2021-10-02T11:56:44.681549Z","iopub.status.idle":"2021-10-02T11:56:44.708505Z","shell.execute_reply.started":"2021-10-02T11:56:44.681511Z","shell.execute_reply":"2021-10-02T11:56:44.707780Z"},"trusted":true},"execution_count":16,"outputs":[]},{"cell_type":"code","source":"down_stack.summary()","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:44.709887Z","iopub.execute_input":"2021-10-02T11:56:44.710144Z","iopub.status.idle":"2021-10-02T11:56:44.948310Z","shell.execute_reply.started":"2021-10-02T11:56:44.710111Z","shell.execute_reply":"2021-10-02T11:56:44.947526Z"},"trusted":true},"execution_count":17,"outputs":[]},{"cell_type":"code","source":"def upsample(filters, size, norm_type='batchnorm', apply_dropout=False):\n\n  initializer = tf.random_normal_initializer(0., 0.02)\n\n  result = tf.keras.Sequential()\n  result.add(\n      tf.keras.layers.Conv2DTranspose(filters, size, strides=2,\n                                      padding='same',\n                                      kernel_initializer=initializer,\n                                      use_bias=False))\n\n  if norm_type.lower() == 'batchnorm':\n    result.add(tf.keras.layers.BatchNormalization())\n  elif norm_type.lower() == 'instancenorm':\n    result.add(InstanceNormalization())\n\n  if apply_dropout:\n    result.add(tf.keras.layers.Dropout(0.5))\n\n  result.add(tf.keras.layers.ReLU())\n\n  return result","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:44.949606Z","iopub.execute_input":"2021-10-02T11:56:44.949896Z","iopub.status.idle":"2021-10-02T11:56:44.959527Z","shell.execute_reply.started":"2021-10-02T11:56:44.949861Z","shell.execute_reply":"2021-10-02T11:56:44.958728Z"},"trusted":true},"execution_count":18,"outputs":[]},{"cell_type":"code","source":"up_stack = [\n    upsample(512, 3),  # 4x4 -> 8x8\n    upsample(256, 3),  # 8x8 -> 16x16\n    upsample(128, 3),  # 16x16 -> 32x32\n    upsample(64, 3),   # 32x32 -> 64x64\n]","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:44.960968Z","iopub.execute_input":"2021-10-02T11:56:44.961367Z","iopub.status.idle":"2021-10-02T11:56:44.996333Z","shell.execute_reply.started":"2021-10-02T11:56:44.961328Z","shell.execute_reply":"2021-10-02T11:56:44.995579Z"},"trusted":true},"execution_count":19,"outputs":[]},{"cell_type":"code","source":"def unet_model(output_channels:int):\n  inputs = tf.keras.layers.Input(shape=[128, 128, 3])    # 128,128\n\n  # Downsampling through the model\n  skips = down_stack(inputs)\n  x = skips[-1]\n  skips = reversed(skips[:-1])\n\n  # Upsampling and establishing the skip connections\n  for up, skip in zip(up_stack, skips):\n    x = up(x)\n    concat = tf.keras.layers.Concatenate()\n    x = concat([x, skip])\n\n  # This is the last layer of the model\n  last = tf.keras.layers.Conv2DTranspose(\n      filters=output_channels, kernel_size=3, strides=2,\n      padding='same')  #64x64 -> 128x128\n\n  x = last(x)\n\n  return tf.keras.Model(inputs=inputs, outputs=x)","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:44.997519Z","iopub.execute_input":"2021-10-02T11:56:44.998299Z","iopub.status.idle":"2021-10-02T11:56:45.006841Z","shell.execute_reply.started":"2021-10-02T11:56:44.998259Z","shell.execute_reply":"2021-10-02T11:56:45.005709Z"},"trusted":true},"execution_count":20,"outputs":[]},{"cell_type":"code","source":"OUTPUT_CLASSES = 22\n\nmodel = unet_model(output_channels=OUTPUT_CLASSES)\nmodel.compile(optimizer= tf.keras.optimizers.Adam(learning_rate=1e-4) ,\n              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n              metrics=['accuracy'])","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:56:45.008237Z","iopub.execute_input":"2021-10-02T11:56:45.008541Z","iopub.status.idle":"2021-10-02T11:56:45.490039Z","shell.execute_reply.started":"2021-10-02T11:56:45.008504Z","shell.execute_reply":"2021-10-02T11:56:45.489174Z"},"trusted":true},"execution_count":21,"outputs":[]},{"cell_type":"code","source":"#+++","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:57:10.236692Z","iopub.execute_input":"2021-10-02T11:57:10.237310Z","iopub.status.idle":"2021-10-02T11:57:10.242363Z","shell.execute_reply.started":"2021-10-02T11:57:10.237274Z","shell.execute_reply":"2021-10-02T11:57:10.241363Z"},"trusted":true},"execution_count":23,"outputs":[]},{"cell_type":"code","source":"def create_mask(pred_mask):\n  pred_mask = tf.argmax(pred_mask, axis=-1)\n  pred_mask = pred_mask[..., tf.newaxis]\n  return pred_mask[0]","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:58:11.070363Z","iopub.execute_input":"2021-10-02T11:58:11.070659Z","iopub.status.idle":"2021-10-02T11:58:11.076042Z","shell.execute_reply.started":"2021-10-02T11:58:11.070630Z","shell.execute_reply":"2021-10-02T11:58:11.075026Z"},"trusted":true},"execution_count":24,"outputs":[]},{"cell_type":"code","source":"class DisplayCallback(tf.keras.callbacks.Callback):\n    def on_epoch_end(self, epoch, logs=None):\n        clear_output(wait=True)\n        show_predictions()\n        print ('\\nSample Prediction after epoch {}\\n'.format(epoch+1))","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:58:12.239983Z","iopub.execute_input":"2021-10-02T11:58:12.240534Z","iopub.status.idle":"2021-10-02T11:58:12.245607Z","shell.execute_reply.started":"2021-10-02T11:58:12.240498Z","shell.execute_reply":"2021-10-02T11:58:12.244654Z"},"trusted":true},"execution_count":25,"outputs":[]},{"cell_type":"code","source":"EPOCHS = 30\nVAL_SUBSPLITS = 5\nVALIDATION_STEPS =  val_size//BATCH_SIZE//VAL_SUBSPLITS\n\nmodel_history = model.fit(train_dataset, epochs=EPOCHS,\n                          steps_per_epoch=STEPS_PER_EPOCH,\n                          validation_steps=VALIDATION_STEPS,\n                          validation_data= val_dataset ,\n                          #callbacks=[DisplayCallback()]\n                         )","metadata":{"execution":{"iopub.status.busy":"2021-10-02T11:58:13.780109Z","iopub.execute_input":"2021-10-02T11:58:13.780710Z","iopub.status.idle":"2021-10-02T12:00:01.366384Z","shell.execute_reply.started":"2021-10-02T11:58:13.780673Z","shell.execute_reply":"2021-10-02T12:00:01.365582Z"},"trusted":true},"execution_count":26,"outputs":[]},{"cell_type":"code","source":"def show_predictions(dataset=None, num=1):\n  if dataset:\n    for image, mask in dataset.take(num):\n      pred_mask = model.predict(image)\n      display([image[0], mask[0], create_mask(pred_mask)])\n  else:\n    display([sample_image, sample_mask,\n             create_mask(model.predict(sample_image[tf.newaxis, ...]))])","metadata":{"execution":{"iopub.status.busy":"2021-10-02T12:02:12.045946Z","iopub.execute_input":"2021-10-02T12:02:12.048016Z","iopub.status.idle":"2021-10-02T12:02:12.053527Z","shell.execute_reply.started":"2021-10-02T12:02:12.047972Z","shell.execute_reply":"2021-10-02T12:02:12.052796Z"},"trusted":true},"execution_count":27,"outputs":[]},{"cell_type":"code","source":"def display(display_list):\n    plt.figure(figsize=(15, 15))\n\n    title = ['Input Image', 'True Mask', 'Predicted Mask']\n\n    for i in range(len(display_list)):\n        plt.subplot(1, len(display_list), i+1)\n        plt.title(title[i])\n        plt.imshow(np.array(display_list[i]))\n        plt.axis('off')\n    plt.show()","metadata":{"execution":{"iopub.status.busy":"2021-10-02T12:02:13.387807Z","iopub.execute_input":"2021-10-02T12:02:13.388069Z","iopub.status.idle":"2021-10-02T12:02:13.396059Z","shell.execute_reply.started":"2021-10-02T12:02:13.388041Z","shell.execute_reply":"2021-10-02T12:02:13.394089Z"},"trusted":true},"execution_count":28,"outputs":[]},{"cell_type":"code","source":"loss = model_history.history['loss']\nval_loss = model_history.history['val_loss']\n\nepochs = range(EPOCHS)\n\nplt.figure()\nplt.plot(epochs, loss, 'r', label='Training loss')\nplt.plot(epochs, val_loss, 'bo', label='Validation loss')\nplt.title('Training and Validation Loss')\nplt.xlabel('Epoch')\nplt.ylabel('Loss Value')\nplt.ylim([0, 1])\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-10-02T12:02:16.998473Z","iopub.execute_input":"2021-10-02T12:02:16.999032Z","iopub.status.idle":"2021-10-02T12:02:17.407864Z","shell.execute_reply.started":"2021-10-02T12:02:16.998988Z","shell.execute_reply":"2021-10-02T12:02:17.407181Z"},"trusted":true},"execution_count":29,"outputs":[]},{"cell_type":"code","source":"show_predictions(train_dataset, num =20)","metadata":{"execution":{"iopub.status.busy":"2021-10-02T12:02:22.567261Z","iopub.execute_input":"2021-10-02T12:02:22.568015Z","iopub.status.idle":"2021-10-02T12:02:32.769714Z","shell.execute_reply.started":"2021-10-02T12:02:22.567976Z","shell.execute_reply":"2021-10-02T12:02:32.769049Z"},"trusted":true},"execution_count":30,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}