\n
Submission File
\n
For each image in the test set, you must predict a list of instance segmentation masks and their associated detection score (Confidence). The submission.csv file uses the following format:
\n
id,height,width,prediction_string\n72e40acccadf,512,512,0 1.0 eNoLTDAwyrM3yI/PMwcAE94DZA==\n
\n
where prediction_string has the format 0 {confidence} {EncodedMask}. Note that the metric has several \"boilerplate\" values needed to adapt it to this competition; namely, the height, width, and the leading 0 in prediction_string, which ordinarily is a class label.
\n
Separate prediction strings multiple instance masks for the same image with a space, like so:
\n
id,height,width,prediction_string\n72e40acccadf,512,512,0 1.0 eNoLTDAwyrM3yI/PMwcAE94DZA== 0 0.5 eAndnnDS1A/mdmkE35Ek9d\n
\n
The binary segmentation masks are run-length encoded (RLE), zlib compressed, and base64 encoded to be used in text format as EncodedMask. Specifically, we use the Coco masks RLE encoding/decoding (see the encode method of COCOโs mask API), the zlib compression/decompression (RFC1950), and vanilla base64 encoding.
","metadata":{}},{"cell_type":"markdown","source":"
\n Tile meta\n
","metadata":{}},{"cell_type":"markdown","source":"
\n
The competition data includes small pieces called \"tiles\" that come from five big pictures called \"Whole Slide Images\" (WSI). These WSIs are split into two groups called datasets. In Dataset 1, the tiles have been looked at by experts who reviewed and marked them. In Dataset 2, the tiles are from the same big pictures but they don't have as many marks, and the marks they do have haven't been reviewed by experts.
\n
","metadata":{}},{"cell_type":"code","source":"tile_meta","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:26.614654Z","iopub.execute_input":"2023-08-11T12:21:26.615432Z","iopub.status.idle":"2023-08-11T12:21:26.629012Z","shell.execute_reply.started":"2023-08-11T12:21:26.615395Z","shell.execute_reply":"2023-08-11T12:21:26.628171Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dataset_count = tile_meta['dataset'].value_counts()\n\nplt.stackplot(list(map(str, dataset_count.index)), dataset_count.values)\n\nplt.xlabel('Dataset Number')\nplt.ylabel('Number of Examples')\nplt.title('Number of examples in each dataset')\n\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:29.674922Z","iopub.execute_input":"2023-08-11T12:21:29.675594Z","iopub.status.idle":"2023-08-11T12:21:29.971210Z","shell.execute_reply.started":"2023-08-11T12:21:29.675556Z","shell.execute_reply":"2023-08-11T12:21:29.969765Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Wsi meta\n
","metadata":{}},{"cell_type":"markdown","source":"
\n
Whole Slide Images (WSIs), also known as virtual slides or digital slides, refer to high-resolution digital representations of entire histopathology glass slides. These slides are typically generated by scanning glass slides using specialized slide scanners.
\n
We should expect 14 unique sources of WSIs where 5 have been used for annotations by either a expert or non expert which then get put into dataset 1 and 2 respectively, and the other 9 correspond to WSIs for dataset 3 which have no annotations. Although we should expect 14 we see that there is only thirteen. A keen eye would see that number 5 is missing. This is probably because the 5th WSI belongs to the first dataset and was removed from the dataset to be used as the test set.
\n
","metadata":{}},{"cell_type":"markdown","source":"## Distribution of WSI by datasets\n","metadata":{}},{"cell_type":"markdown","source":"## Legend \n","metadata":{}},{"cell_type":"code","source":"wsi_meta","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:35.294818Z","iopub.execute_input":"2023-08-11T12:21:35.295309Z","iopub.status.idle":"2023-08-11T12:21:35.313400Z","shell.execute_reply.started":"2023-08-11T12:21:35.295265Z","shell.execute_reply":"2023-08-11T12:21:35.311834Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sex_counts = wsi_meta.sex.value_counts()\nrace_counts = wsi_meta.race.value_counts()\n\nfig, (ax1, ax2) = plt.subplots(1, 2)\n\nax1.pie(sex_counts.values, labels=sex_counts.index, autopct='%1.1f%%')\nax1.set_title(\"Distribution of Sexes\")\n\nax2.pie(race_counts.values, labels=race_counts.index, autopct='%1.1f%%')\nax2.set_title(\"Distribution of races\")\n\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:37.335413Z","iopub.execute_input":"2023-08-11T12:21:37.335891Z","iopub.status.idle":"2023-08-11T12:21:37.719403Z","shell.execute_reply.started":"2023-08-11T12:21:37.335857Z","shell.execute_reply":"2023-08-11T12:21:37.717593Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sex_counts = wsi_meta.sex.value_counts()\nrace_counts = wsi_meta.race.value_counts()\n\nfig, (ax1, ax2) = plt.subplots(1, 2)\n\nax1.specgram(sex_counts.values)\nax1.set_title(\"Distribution of Sexes\")\n\n# ax2.pie(race_counts.values, labels=race_counts.index, autopct='%1.1f%%')\n# ax2.set_title(\"Distribution of races\")\n\nax2.specgram(race_counts.values)\nax2.set_title(\"Distribution of races\")\n\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:39.616572Z","iopub.execute_input":"2023-08-11T12:21:39.617073Z","iopub.status.idle":"2023-08-11T12:21:40.063330Z","shell.execute_reply.started":"2023-08-11T12:21:39.617040Z","shell.execute_reply":"2023-08-11T12:21:40.062100Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sex_counts = wsi_meta.sex.value_counts()\nrace_counts = wsi_meta.race.value_counts()\n\nfig, (ax1, ax2) = plt.subplots(1, 2)\n\nax1.violinplot(sex_counts.values)\nax1.set_title(\"Distribution of Sexes\")\n\nax2.violinplot(race_counts.values)\nax2.set_title(\"Distribution of races\")\n\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:41.605994Z","iopub.execute_input":"2023-08-11T12:21:41.607437Z","iopub.status.idle":"2023-08-11T12:21:42.122898Z","shell.execute_reply.started":"2023-08-11T12:21:41.607396Z","shell.execute_reply":"2023-08-11T12:21:42.122006Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Polygons-annotation ๐ฏ\n
","metadata":{}},{"cell_type":"markdown","source":"****\n","metadata":{}},{"cell_type":"markdown","source":"
\nPolygonal segmentation masks in JSONL format, available for Dataset 1 and Dataset 2. Each line gives JSON annotations for a single image with:
\nid Identifies the corresponding image in train/annotations A list of mask annotations with:type Identifies the type of structure annotated:\nblood_vessel The target structure. Your goal in this competition is to predict these kinds of masks on the test set.glomerulus A capillary ball structure in the kidney. These parts of the images were excluded from blood vessel annotation. You should ensure none of your test set predictions occur within glomerulus structures as they will be counted as false positives. Annotations are provided for test set tiles.unsure A structure the expert annotators cannot confidently distinguish as a blood vessel.
coordinates A list of polygon coordinates defining the segmentation mask.
\n
\n\n****","metadata":{}},{"cell_type":"markdown","source":"
\n Setup Pipeline โ๏ธ\n
","metadata":{}},{"cell_type":"markdown","source":"
\n
pycocotools to encode masks and their subsequent evaluation.
\n Set a random number generator for all frameworks.\n
","metadata":{}},{"cell_type":"code","source":"import torch\nfrom torch import nn\nimport random\n\nimport shutil\nimport os\nimport sys\nfrom colorama import Fore","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:45.235514Z","iopub.execute_input":"2023-08-11T12:21:45.236233Z","iopub.status.idle":"2023-08-11T12:21:49.058390Z","shell.execute_reply.started":"2023-08-11T12:21:45.236194Z","shell.execute_reply":"2023-08-11T12:21:49.057030Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Class SetupPipline has three methods:\n\n - __pycocotools(): This method is a static method, which means that it can be called without creating an instance of the class. This method is used to install the PyCocoTools library, which is a Python library for working with COCO datasets.\n - seed_everything(): This method is also a static method. It is used to set the random seed for various Python libraries, such as NumPy, PyTorch, and CuDNN. This helps to ensure that the results of the code are reproducible.\n - __call__(): This is the main method of the SetupPipline class. It is called when an instance of the class is created. This method takes two arguments: seed and pycoco. The seed argument is the random seed to use, and the pycoco argument is a boolean value that indicates whether to install the PyCocoTools library.\n \n \n ### Details\n- __pycocotools(): This method first checks if the /kaggle/working/packages directory exists. If it does not exist, the method creates it. The method then copies the pycocotools directory from the /kaggle/input/hubmap-tools-ultralytics-and-pycocotools directory to the /kaggle/working/packages directory. The method then changes the working directory to the /kaggle/working/packages/pycocotools-2.0.6 directory and runs the following commands:\n - python setup.py install\n - pip install . --no-index --find-links /kaggle/working/packages/ The first command installs the PyCocoTools library from the source code. The second command installs the PyCocoTools library from the /kaggle/working/packages directory.\n- seed_everything(): This method sets the random seed for the following Python libraries:\n - random\n - np.random\n - torch.manual_seed\n - torch.cuda.manual_seed\n - torch.backends.cudnn.deterministic The random seed is set to the value of the seed argument.\n- __call__(): This method takes two arguments: seed and pycoco. The seed argument is the random seed to use, and the pycoco argument is a boolean value that indicates whether to install the PyCocoTools library.\n= If the pycoco argument is True, the __pycocotools() method is called to install the PyCocoTools library. If the seed argument is not None, the seed_everything() method is called to set the random seed.","metadata":{}},{"cell_type":"code","source":"class SetupPipline: \n @staticmethod\n def __pycocotools() -> None:\n if not os.path.exists(\"/kaggle/working/packages\"):\n shutil.copytree(\"/kaggle/input/hubmap-tools-ultralytics-and-pycocotools/pycocotools/pycocotools\", \"/kaggle/working/packages\")\n os.chdir(\"/kaggle/working/packages/pycocotools-2.0.6/\")\n os.system(\"python setup.py install\")\n os.system(\"pip install . --no-index --find-links /kaggle/working/packages/\")\n os.chdir(\"/kaggle/working\")\n\n @staticmethod\n def seed_everything(seed: int) -> None:\n random.seed(seed)\n np.random.seed(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.backends.cudnn.deterministic = True\n \n def __call__(self, seed: int = 42, pycoco: bool = True) -> None:\n if pycoco:\n self.__pycocotools()\n if seed:\n self.seed_everything(seed)","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:21:49.061384Z","iopub.execute_input":"2023-08-11T12:21:49.062428Z","iopub.status.idle":"2023-08-11T12:21:49.073027Z","shell.execute_reply.started":"2023-08-11T12:21:49.062374Z","shell.execute_reply":"2023-08-11T12:21:49.071895Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%capture\nsetup = SetupPipline()\nSEED: int = 42\nsetup(seed=SEED, pycoco=True)","metadata":{"_kg_hide-input":false,"_kg_hide-output":true,"execution":{"iopub.status.busy":"2023-08-11T12:21:53.976310Z","iopub.execute_input":"2023-08-11T12:21:53.976779Z","iopub.status.idle":"2023-08-11T12:22:57.156922Z","shell.execute_reply.started":"2023-08-11T12:21:53.976745Z","shell.execute_reply":"2023-08-11T12:22:57.155554Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Dataset libraries\n
","metadata":{}},{"cell_type":"code","source":"from torch.utils.data import Dataset\nimport cv2\nimport yaml\nimport json\nfrom PIL import Image","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:11.513992Z","iopub.execute_input":"2023-08-11T12:23:11.514463Z","iopub.status.idle":"2023-08-11T12:23:11.720591Z","shell.execute_reply.started":"2023-08-11T12:23:11.514429Z","shell.execute_reply":"2023-08-11T12:23:11.719338Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Dataset config\n
","metadata":{}},{"cell_type":"code","source":"dataset_config = {\n \"background\": {\n \"apply_mask\": None,\n \"label\": 0,\n \"rgb\": (0, 0, 0),\n \"loss_weight\": None\n },\n \"blood_vessel\": {\n \"apply_mask\": True,\n \"label\": 1,\n \"rgb\": (255, 8, 8),\n \"loss_weight\": None\n },\n \"glomerulus\": {\n \"apply_mask\": True,\n \"label\": 2,\n \"rgb\": (8, 12, 255),\n \"loss_weight\": None\n },\n \"unsure\": {\n \"apply_mask\": True,\n \"label\": 3,\n \"rgb\": (8, 255, 20),\n \"loss_weight\": None\n }\n}","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:14.187255Z","iopub.execute_input":"2023-08-11T12:23:14.187762Z","iopub.status.idle":"2023-08-11T12:23:14.196798Z","shell.execute_reply.started":"2023-08-11T12:23:14.187727Z","shell.execute_reply":"2023-08-11T12:23:14.195247Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"- write_config function takes two arguments: data and path. The data argument is a dictionary of dictionaries. The path argument is the path to the file to write the configuration to.\n- The write_config function first opens the file at the specified path in write mode. Then, it uses the yaml.safe_dump() function to write the configuration data to the file. The yaml.safe_dump() function takes two arguments:\n - stream: The file object to write the data to.\n - data: The data to write to the file.\n \n- yaml.safe_dump() function will try to pretty-print the configuration data, but it will not if the data contains any non-UTF8 characters.","metadata":{}},{"cell_type":"code","source":"def write_config(data: dict[dict, ...], path: str) -> None:\n with open(path, mode=\"w\") as f:\n yaml.safe_dump(stream=f, data=data)\n \nwrite_config(dataset_config, __DATASET_CONFIG_PATH)","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:17.034769Z","iopub.execute_input":"2023-08-11T12:23:17.035848Z","iopub.status.idle":"2023-08-11T12:23:17.045708Z","shell.execute_reply.started":"2023-08-11T12:23:17.035810Z","shell.execute_reply":"2023-08-11T12:23:17.044111Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Dataset ๐ชข\n
","metadata":{}},{"cell_type":"markdown","source":"### HuBMAPDataset class has following methods:\n\n- __init__(): This method is the constructor of the class. It takes three arguments: annotation_path, image_path, and config_path. The annotation_path argument is the path to the file containing the annotations for the dataset. The image_path argument is the path to the directory containing the images for the dataset. The config_path argument is the path to the file containing the configuration for the dataset.\n- __len__(): This method returns the length of the dataset.\n- __getitem__(idx): This method returns the idxth sample from the dataset. The sample is a tuple of two elements: the image and the mask. The image is a NumPy array of shape (512, 512, 3). The mask is a NumPy array of shape (512, 512, 1).\n\n- HuBMAPDataset class first parses the annotations file to get a list of samples. Each sample is a dictionary that contains the following keys:\n - id: The ID of the sample.\n - annotations: A list of annotations for the sample.\n \n- HuBMAPDataset loads the image and mask for each sample. The image is loaded from the file system and converted to a NumPy array. The mask is created by filling a NumPy array with zeros. For each annotation in the sample, the coordinates of the annotation are used to fill the mask with the color specified in the configuration file.Finally, the HuBMAPDataset class returns a tuple of the image and the mask.","metadata":{}},{"cell_type":"code","source":"class HuBMAPDataset(Dataset):\n def __init__(self, \n annotation_path: str,\n image_path: str, \n config_path: str):\n self.__image_path = image_path\n self.__samples = self.parse_jsonl(annotation_path)\n self.__config = self.load_config(config_path)\n \n def __len__(self) -> int:\n return len(self.__samples)\n\n def __getitem__(self, idx: int) -> tuple[np.ndarray, np.ndarray]:\n image = self.__get_image(idx)\n mask = self.__get_mask(idx)\n # image = torch.tensor(image, dtype=torch.float32).permute(2, 0, 1)\n # mask = torch.tensor(mask, dtype=torch.float32) \n return image, mask\n \n @staticmethod\n def parse_jsonl(path: str) -> list[dict, ...]:\n with open(path, 'r') as json_file:\n jsonl_labels = [\n json.loads(line)\n for line in notebook.tqdm(\n json_file, desc=\"Processing polygons\", total=1633\n )\n ]\n return jsonl_labels\n\n @staticmethod\n def load_config(path: str) -> dict:\n with open(path, mode=\"r\") as f:\n data = yaml.load(stream=f, Loader=yaml.SafeLoader)\n return data\n \n def __get_image_path(self, id: str) -> str:\n path = os.path.join(\n self.__image_path, f\"{id}.tif\"\n )\n return path\n \n def __get_image(self, idx: int) -> np.ndarray:\n id = self.__samples[idx][\"id\"]\n image_path = self.__get_image_path(id)\n image = Image.open(image_path)\n image = np.asarray(image)\n return image\n \n def __get_mask(self, idx: int) -> np.ndarray:\n mask = np.zeros((512, 512), dtype=np.uint8)\n annotations = self.__samples[idx][\"annotations\"]\n \n for vessel in annotations:\n vessel_type = vessel[\"type\"] \n config = self.__config[vessel_type]\n \n if config[\"apply_mask\"]:\n coordinates = np.array(vessel[\"coordinates\"])\n mask = cv2.fillPoly(\n mask, pts=coordinates,\n color=config[\"rgb\"]\n )\n return mask","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:21.147021Z","iopub.execute_input":"2023-08-11T12:23:21.148519Z","iopub.status.idle":"2023-08-11T12:23:21.163643Z","shell.execute_reply.started":"2023-08-11T12:23:21.148470Z","shell.execute_reply":"2023-08-11T12:23:21.162811Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Let's check dataset\n
","metadata":{}},{"cell_type":"code","source":"dataset = HuBMAPDataset(__ANNOTATION_PATH, __TRAIN_PATH, __DATASET_CONFIG_PATH)","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:24.174804Z","iopub.execute_input":"2023-08-11T12:23:24.175806Z","iopub.status.idle":"2023-08-11T12:23:30.892575Z","shell.execute_reply.started":"2023-08-11T12:23:24.175760Z","shell.execute_reply":"2023-08-11T12:23:30.891370Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"image, mask = dataset[0]\n\nfig, (ax1, ax2) = plt.subplots(1, 2)\n\nax1.imshow(image)\nax2.imshow(mask)\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:37.495762Z","iopub.execute_input":"2023-08-11T12:23:37.496213Z","iopub.status.idle":"2023-08-11T12:23:38.054738Z","shell.execute_reply.started":"2023-08-11T12:23:37.496180Z","shell.execute_reply":"2023-08-11T12:23:38.053553Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Submission ๐\n
\n\n
\n
id, height, width, prediction_string\n
\n - id - identifier of the image that came to the input of the neural network.
\n - height - image height (constant value equal to 512).
\n - width - image width (constant value equal to 512).
\n - prediction_string:\n
\n - label - label class constant value equal to 0 (at least as I understand it, correct me if I'm wrong)
\n - confidence - the confidence of the model that this object or instance belongs to the target class. That is, your model will give a mask at the output, where each mask will have the probability that this particular mask belongs to the desired class. It's like the probability of a model belonging to an object to a class. In fact, this is a probabilistic membership score that always appears when it comes to classification.
\n
\n \n
\n
\n\n","metadata":{}},{"cell_type":"markdown","source":"
\n Important ๐\n
I removed the full implementation so that you understand that you must override the forward method, since I do not know what exactly your neural network returns and this could lead to confusion. In general, the class turned out to be quite comfortable. You only need to pass the model and the path to the test directory, and then call the submit() method. Please let me know if something doesn't work!
\n
","metadata":{}},{"cell_type":"code","source":"import base64\nimport numpy as np\nimport torch\nimport torch.nn as nn\nfrom pycocotools import _mask as coco_mask\nimport typing as t\nimport zlib\nimport pandas as pd\nimport torchvision.transforms as T","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:43.644965Z","iopub.execute_input":"2023-08-11T12:23:43.645410Z","iopub.status.idle":"2023-08-11T12:23:43.934573Z","shell.execute_reply.started":"2023-08-11T12:23:43.645379Z","shell.execute_reply":"2023-08-11T12:23:43.933366Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# EncodeBinaryMask class has the following methods:\n\n- __checking_mask(): This method checks if the mask is a binary mask. If the mask is not a binary mask, the method raises an error.\n- __convert_mask(): This method converts the mask to a format that can be encoded by the coco_mask library.\n- __compress_encode(): This method compresses the encoded mask and encodes it in base64 format.\n- __call__(): This is the main method of the class. It takes one argument: mask. The mask argument is the binary mask to encode.\n\n- __call__() method first checks the mask to make sure it is a binary mask. If the mask is not a binary mask, the method raises an error.\n\n- __call__() method then converts the mask to a format that can be encoded by the coco_mask library. The coco_mask library is a Python library for working with masks in the COCO format.\n\n- __call__() method then encodes the mask and compresses it in base64 format. The base64 format is a standard way to encode binary data in a text format.\n\n- __call__() method finally returns the base64 encoded mask.","metadata":{}},{"cell_type":"code","source":"class EncodeBinaryMask:\n @staticmethod\n def __checking_mask(mask: np.ndarray) -> np.ndarray:\n if mask.dtype != np.bool:\n raise ValueError(\n \"expects a binary mask, received dtype == %s\" %\n mask.dtype\n )\n return mask\n\n @staticmethod\n def __convert_mask(mask: np.ndarray):\n mask_to_encode = mask.astype(np.uint8)\n mask_to_encode = np.asfortranarray(mask_to_encode)\n return mask_to_encode\n\n @staticmethod\n def __compress_encode(encoded_mask) -> t.Text:\n binary_str = zlib.compress(encoded_mask, zlib.Z_BEST_COMPRESSION)\n base64_str = base64.b64encode(binary_str)\n return base64_str\n\n def __call__(self, mask: np.ndarray) -> t.Text:\n mask = self.__checking_mask(mask)\n mask_to_encode = self.__convert_mask(mask)\n encoded_mask = coco_mask.encode(mask_to_encode)[0][\"counts\"]\n base64_str = self.__compress_encode(encoded_mask)\n return base64_str\n","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:23:51.406006Z","iopub.execute_input":"2023-08-11T12:23:51.407098Z","iopub.status.idle":"2023-08-11T12:23:51.417073Z","shell.execute_reply.started":"2023-08-11T12:23:51.407058Z","shell.execute_reply":"2023-08-11T12:23:51.415823Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Submission:\n def __init__(self, dirpath: str, model: torch.nn.Module):\n self.__eval_transforms = self.get_transforms()\n self.__model = model\n self.__encoder = EncodeBinaryMask()\n self.__dirpath = dirpath\n self.__filenames = os.listdir(dirpath)\n self.conv1 = nn.Conv2d(3,512,512)\n self.pool = nn.MaxPool2d(1, 1)\n self.conv2 = nn.Conv2d(3,512,512)\n self.fc = nn.Linear(512,10)\n \n self.__submission_dict = {\n \"id\": [],\n \"height\": [],\n \"width\": [],\n \"prediction_string\": []\n }\n \n self.submission = None\n \n @staticmethod\n def get_transforms():\n return T.Compose([\n T.ToTensor(),\n T.Resize(size=(512, 512)),\n T.Normalize(mean=[0.485, 0.456, 0.406],\n std=[0.229, 0.224, 0.225])\n ])\n\n def __len__(self):\n return len(self.__filenames)\n\n def __get_columns(self) -> None:\n for filename in self.__filenames:\n path = self.__get_image_path(filename)\n image = self.__get_image(path)\n masks = self.__forward(image)\n identifier, height, width, prediction_string = self.__get_cells(filename, masks)\n self.__update_columns(identifier, height, width, prediction_string)\n\n def __update_columns(self, identifier: str, height: int, width: int, prediction_string: str) -> None:\n self.__submission_dict[\"id\"].append(identifier)\n self.__submission_dict[\"height\"].append(height)\n self.__submission_dict[\"width\"].append(width)\n self.__submission_dict[\"prediction_string\"].append(prediction_string)\n\n def __get_cells(self, filename: str, masks: list):\n prediction_string = \"\"\n prediction_string = self.__get_prediction_string(masks, prediction_string)\n identifier = filename.split(\".\")[0]\n height, width = mask.shape\n return identifier, height, width, prediction_string\n\n def __get_prediction_string(self, masks: list, prediction_string: str) -> str:\n \"\"\"\n If the neural network did not find the target structure,\n then we will return an empty prediction_string.\n \n However, the columns:\n id, height, width - must be present for all test images!!!\n \"\"\"\n \n if masks:\n for outputs in masks:\n mask = outputs[\"mask\"].detach().permute(1,2,0).cpu().numpy()\n mask = np.where(mask > 0.5, 1, 0).astype(np.bool)\n base64_str = self.__encoder(mask)\n confidence = outputs[\"confidence\"]\n prediction_string += f\"0 {confidence} {base64_str.decode('utf-8')} \"\n return prediction_string\n\n def __get_image_path(self, filename: str) -> str:\n return os.path.join(\n self.__dirpath, filename\n )\n\n def __get_image(self, path: str) -> torch.Tensor:\n image = Image.open(path)\n image = np.asarray(image)\n image = self.__eval_transforms(image)\n return image\n\n # You must implement this function to work with your custom neural network!\n def __forward(self, image: torch.tensor) ->list:\n \"\"\"\n This fuction should return list with masks & confidence.\n And each mask we shoulde encode (see EncodeBinaryMask & sample_submission file) \n \n outputs <- model(image) \n outputs -> [\n {\"mask\": mask1, \"confidence\": confidence1},\n {\"mask\": mask2, \"confidence\": confidence2}, \n ...,\n {\"mask\": maskN, \"confidence\": confidenceN}\n ]\n \n Example:\n \"\"\"\n masks = self.__model(image) # Mask must have shape 1x512x512\n masks = torch.relu(image)\n masks= self.pool(nn.functional.relu(self.conv1(image)))\n masks = self.pool(nn.functional.relu(self.conv2(image)))\n masks = torch.flatten(image)\n masks = torch.relu(image)\n masks = torch.log_softmax(image,dim=0)\n masks = nn.functional.log_softmax(self.fc(image),dim=1)\n mask1 = torch.nn.Flatten(0, 1)\n masks = torch.nn.Flatten(mask1(image))\n masks = torch.nn.Flatten(0, 1)\n# masks = torch.sum(image)\n masks = self.__model(image) * 5\n return masks \n\n def submit(self) -> None:\n if not self.submission:\n self.__get_columns()\n self.submission = pd.DataFrame(self.__submission_dict)\n self.submission = self.submission.set_index('id')\n self.submission.to_csv(\"submission.csv\")\n","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:58:16.033294Z","iopub.execute_input":"2023-08-11T12:58:16.033703Z","iopub.status.idle":"2023-08-11T12:58:16.062110Z","shell.execute_reply.started":"2023-08-11T12:58:16.033658Z","shell.execute_reply":"2023-08-11T12:58:16.060710Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n Sample model\n
","metadata":{}},{"cell_type":"code","source":"class MyBestModel:\n @staticmethod\n def generate_masks(num_masks: int) -> list[dict, ...]:\n masks = []\n for _ in range(num_masks):\n mask = torch.randint(0, 2, (1, 512, 512))\n confidence = round(float(torch.rand(1)[0]), 2)\n masks.append({\"mask\": mask, \"confidence\": confidence})\n return masks\n \n def __call__(self, image) -> list[dict, ...]:\n num_masks = torch.randint(1, 5, (1, 1))\n masks = self.generate_masks(num_masks)\n return masks","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:56:31.810308Z","iopub.execute_input":"2023-08-11T12:56:31.811607Z","iopub.status.idle":"2023-08-11T12:56:31.820229Z","shell.execute_reply.started":"2023-08-11T12:56:31.811542Z","shell.execute_reply":"2023-08-11T12:56:31.818686Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model = MyBestModel()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:56:35.479134Z","iopub.execute_input":"2023-08-11T12:56:35.479521Z","iopub.status.idle":"2023-08-11T12:56:35.486396Z","shell.execute_reply.started":"2023-08-11T12:56:35.479493Z","shell.execute_reply":"2023-08-11T12:56:35.485293Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sub = Submission(dirpath=__TEST_PATH, model=model)\nsub.submit()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:56:37.480284Z","iopub.execute_input":"2023-08-11T12:56:37.481312Z","iopub.status.idle":"2023-08-11T12:56:49.986753Z","shell.execute_reply.started":"2023-08-11T12:56:37.481263Z","shell.execute_reply":"2023-08-11T12:56:49.984896Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sub.submission.head()","metadata":{"execution":{"iopub.status.busy":"2023-08-11T12:45:38.284487Z","iopub.execute_input":"2023-08-11T12:45:38.284968Z","iopub.status.idle":"2023-08-11T12:45:38.301999Z","shell.execute_reply.started":"2023-08-11T12:45:38.284934Z","shell.execute_reply":"2023-08-11T12:45:38.300611Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}