{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"#

Import Required Libraries 📚

","metadata":{}},{"cell_type":"code","source":"%%writefile infer.py\nimport argparse\nimport os\nimport gc\nimport cv2\nimport math\nimport copy\nimport time\nimport random\nimport glob\nfrom PIL import Image\nfrom matplotlib import pyplot as plt\n\n# For data manipulation\nimport numpy as np\nimport pandas as pd\n\n# Pytorch Imports\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nimport torch.nn.functional as F\nfrom torch.optim import lr_scheduler\nfrom torch.utils.data import Dataset, DataLoader\nfrom torch.cuda import amp\nimport torchvision\n\n# Utils\nimport joblib\nfrom tqdm import tqdm\nfrom collections import defaultdict\n\n# Sklearn Imports\nfrom sklearn.preprocessing import LabelEncoder\nfrom sklearn.model_selection import StratifiedKFold\n\n# For Image Models\nimport timm\n\n# Albumentations for augmentations\nimport albumentations as A\nfrom albumentations.pytorch import ToTensorV2\n\n# For colored terminal text\nfrom colorama import Fore, Back, Style\nb_ = Fore.BLUE\nsr_ = Style.RESET_ALL\n\nimport warnings\nwarnings.filterwarnings(\"ignore\")\n\n# For descriptive error messages\nos.environ['CUDA_LAUNCH_BLOCKING'] = \"1\"","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:07.352436Z","iopub.execute_input":"2023-10-28T02:38:07.352797Z","iopub.status.idle":"2023-10-28T02:38:07.3658Z","shell.execute_reply.started":"2023-10-28T02:38:07.352769Z","shell.execute_reply":"2023-10-28T02:38:07.36491Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\nparser = argparse.ArgumentParser()\nparser.add_argument('--imgsize', type=int, default=2048)\nparser.add_argument('--batchsize', type=int, default=4)\nparser.add_argument('--weight', type=str, default=None)\nparser.add_argument('--labelpkl', type=str, default=None)\nparser.add_argument('--modelname', type=str, default=\"tf_efficientnet_b0_ns\") # \"convnext_small.in12k_ft_in1k_384\", \"tf_efficientnet_b0_ns\"\nparser.add_argument('--testcsv', type=str, default=None)\nparser.add_argument('--numclasses',type=int, default=5)\n\nargs = parser.parse_args()\nprint(f\"args.imgsize = {args.imgsize}\")\nprint(f\"args.batchsize = {args.batchsize}\")\nprint(f\"args.weight = {args.weight}\")\nprint(f\"args.labelpkl = {args.labelpkl}\")\nprint(f\"args.modelname = {args.modelname}\")\nprint(f\"debug: args.testcsv = {args.testcsv}\")\nprint(f\"debug: args.numclasses = {args.numclasses}\")","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:07.88175Z","iopub.execute_input":"2023-10-28T02:38:07.88213Z","iopub.status.idle":"2023-10-28T02:38:07.889692Z","shell.execute_reply.started":"2023-10-28T02:38:07.882103Z","shell.execute_reply":"2023-10-28T02:38:07.888574Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#

Training Configuration ⚙️

","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\nCONFIG = {\n \"seed\": 42,\n \"img_size\": args.imgsize,\n \"model_name\": args.modelname,\n \"num_classes\": args.numclasses,\n \"valid_batch_size\": args.batchsize,\n \"device\": torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\"),\n}","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:07.892329Z","iopub.execute_input":"2023-10-28T02:38:07.893096Z","iopub.status.idle":"2023-10-28T02:38:07.905553Z","shell.execute_reply.started":"2023-10-28T02:38:07.893059Z","shell.execute_reply":"2023-10-28T02:38:07.903563Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#

Set Seed for Reproducibility

","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\ndef set_seed(seed=42):\n '''Sets the seed of the entire notebook so results are the same every time we run.\n This is for REPRODUCIBILITY.'''\n np.random.seed(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n # When running on the CuDNN backend, two further options must be set\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n # Set a fixed value for the hash seed\n os.environ['PYTHONHASHSEED'] = str(seed)\n \nset_seed(CONFIG['seed'])","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:07.907295Z","iopub.execute_input":"2023-10-28T02:38:07.907908Z","iopub.status.idle":"2023-10-28T02:38:07.914386Z","shell.execute_reply.started":"2023-10-28T02:38:07.907864Z","shell.execute_reply":"2023-10-28T02:38:07.913395Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\nROOT_DIR = '/kaggle/input/UBC-OCEAN'\nTEST_DIR = '/kaggle/input/UBC-OCEAN/test_thumbnails'\n\nLABEL_ENCODER_BIN = args.labelpkl\nBEST_WEIGHT = args.weight","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:08.790894Z","iopub.execute_input":"2023-10-28T02:38:08.791606Z","iopub.status.idle":"2023-10-28T02:38:08.796671Z","shell.execute_reply.started":"2023-10-28T02:38:08.791574Z","shell.execute_reply":"2023-10-28T02:38:08.795824Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\ndef get_test_file_path(image_id):\n return f\"{TEST_DIR}/{image_id}_thumbnail.png\"","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:08.798482Z","iopub.execute_input":"2023-10-28T02:38:08.798776Z","iopub.status.idle":"2023-10-28T02:38:08.806733Z","shell.execute_reply.started":"2023-10-28T02:38:08.798752Z","shell.execute_reply":"2023-10-28T02:38:08.805784Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#

Read the Data 📖

","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\nif args.testcsv is None:\n df = pd.read_csv(f\"{ROOT_DIR}/test.csv\")\nelse:\n df = pd.read_csv(args.testcsv)\n\nif not 'file_path' in df.columns:\n df['file_path'] = df['image_id'].apply(get_test_file_path)\ndf['label'] = 0 # dummy\ndf","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:08.899836Z","iopub.execute_input":"2023-10-28T02:38:08.900108Z","iopub.status.idle":"2023-10-28T02:38:08.905482Z","shell.execute_reply.started":"2023-10-28T02:38:08.900087Z","shell.execute_reply":"2023-10-28T02:38:08.904611Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\ndf_sub = pd.read_csv(f\"{ROOT_DIR}/sample_submission.csv\")\ndf_sub","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:08.90727Z","iopub.execute_input":"2023-10-28T02:38:08.907762Z","iopub.status.idle":"2023-10-28T02:38:08.915478Z","shell.execute_reply.started":"2023-10-28T02:38:08.907738Z","shell.execute_reply":"2023-10-28T02:38:08.914592Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\nencoder = joblib.load( LABEL_ENCODER_BIN )","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:08.91663Z","iopub.execute_input":"2023-10-28T02:38:08.916878Z","iopub.status.idle":"2023-10-28T02:38:08.925488Z","shell.execute_reply.started":"2023-10-28T02:38:08.916856Z","shell.execute_reply":"2023-10-28T02:38:08.924644Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\ndef get_cropped_images(file_path, image_id, th_area = 1000):\n image = Image.open(file_path)\n # Aspect ratio\n as_ratio = image.size[0] / image.size[1]\n \n sxs, exs, sys, eys = [],[],[],[]\n if as_ratio >= 1.5:\n # Crop\n mask = np.max( np.array(image) > 0, axis=-1 ).astype(np.uint8)\n retval, labels = cv2.connectedComponents(mask)\n if retval >= as_ratio:\n x, y = np.meshgrid( np.arange(image.size[0]), np.arange(image.size[1]) )\n for label in range(1, retval):\n area = np.sum(labels == label)\n if area < th_area:\n continue\n xs, ys= x[ labels == label ], y[ labels == label ]\n sx, ex = np.min(xs), np.max(xs)\n cx = (sx + ex) // 2\n crop_size = image.size[1]\n sx = max(0, cx-crop_size//2)\n ex = min(sx + crop_size - 1, image.size[0]-1)\n sx = ex - crop_size + 1\n sy, ey = 0, image.size[1]-1\n sxs.append(sx)\n exs.append(ex)\n sys.append(sy)\n eys.append(ey)\n else:\n crop_size = image.size[1]\n for i in range(int(as_ratio)):\n sxs.append( i * crop_size )\n exs.append( (i+1) * crop_size - 1 )\n sys.append( 0 )\n eys.append( crop_size - 1 )\n else:\n # Not Crop (entire image)\n sxs, exs, sys, eys = [0,],[image.size[0]-1],[0,],[image.size[1]-1]\n\n df_crop = pd.DataFrame()\n df_crop[\"image_id\"] = [image_id] * len(sxs)\n df_crop[\"file_path\"] = [file_path] * len(sxs)\n df_crop[\"sx\"] = sxs\n df_crop[\"ex\"] = exs\n df_crop[\"sy\"] = sys\n df_crop[\"ey\"] = eys\n return df_crop","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:09.480835Z","iopub.execute_input":"2023-10-28T02:38:09.48127Z","iopub.status.idle":"2023-10-28T02:38:09.489298Z","shell.execute_reply.started":"2023-10-28T02:38:09.481235Z","shell.execute_reply":"2023-10-28T02:38:09.488243Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\ndfs = []\nfor (file_path, image_id) in zip(df[\"file_path\"], df[\"image_id\"]):\n dfs.append( get_cropped_images(file_path, image_id) )\n\ndf_crop = pd.concat(dfs)\ndf_crop[\"label\"] = 0 # dummy\ndf_crop","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:12.602368Z","iopub.execute_input":"2023-10-28T02:38:12.602725Z","iopub.status.idle":"2023-10-28T02:38:12.608454Z","shell.execute_reply.started":"2023-10-28T02:38:12.6027Z","shell.execute_reply":"2023-10-28T02:38:12.607547Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\n#df_crop = df_crop.drop_duplicates(subset=[\"image_id\", \"sx\", \"ex\", \"sy\", \"ey\"]).reset_index(drop=True)\n#df_crop","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:12.610017Z","iopub.execute_input":"2023-10-28T02:38:12.610319Z","iopub.status.idle":"2023-10-28T02:38:12.6255Z","shell.execute_reply.started":"2023-10-28T02:38:12.610288Z","shell.execute_reply":"2023-10-28T02:38:12.624631Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#

Dataset Class

","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\nclass UBCDataset(Dataset):\n def __init__(self, df, transforms=None):\n self.df = df\n self.file_names = df['file_path'].values\n self.labels = df['label'].values\n self.transforms = transforms\n self.sxs = df[\"sx\"].values\n self.exs = df[\"ex\"].values\n self.sys = df[\"sy\"].values\n self.eys = df[\"ey\"].values\n \n def __len__(self):\n return len(self.df)\n \n def __getitem__(self, index):\n img_path = self.file_names[index]\n sx = self.sxs[index]\n ex = self.exs[index]\n sy = self.sys[index]\n ey = self.eys[index]\n img = cv2.imread(img_path)\n img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n label = self.labels[index]\n \n img = img[ sy:ey, sx:ex, : ]\n \n if self.transforms:\n img = self.transforms(image=img)[\"image\"]\n \n return {\n 'image': img,\n 'label': torch.tensor(label, dtype=torch.long)\n }","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:13.900789Z","iopub.execute_input":"2023-10-28T02:38:13.901484Z","iopub.status.idle":"2023-10-28T02:38:13.908662Z","shell.execute_reply.started":"2023-10-28T02:38:13.901443Z","shell.execute_reply":"2023-10-28T02:38:13.907821Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#

Augmentations

","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\ndata_transforms = {\n \"valid\": A.Compose([\n A.Resize(CONFIG['img_size'], CONFIG['img_size']),\n A.Normalize(\n mean=[0.485, 0.456, 0.406], \n std=[0.229, 0.224, 0.225], \n max_pixel_value=255.0, \n p=1.0\n ),\n ToTensorV2()], p=1.)\n}","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:13.938649Z","iopub.execute_input":"2023-10-28T02:38:13.939635Z","iopub.status.idle":"2023-10-28T02:38:13.944942Z","shell.execute_reply.started":"2023-10-28T02:38:13.939598Z","shell.execute_reply":"2023-10-28T02:38:13.943977Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#

GeM Pooling

\n\nCode taken from GeM Pooling Explained\n\n![](https://i.imgur.com/thTgYWG.jpg)","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\nclass GeM(nn.Module):\n def __init__(self, p=3, eps=1e-6):\n super(GeM, self).__init__()\n self.p = nn.Parameter(torch.ones(1)*p)\n self.eps = eps\n\n def forward(self, x):\n return self.gem(x, p=self.p, eps=self.eps)\n \n def gem(self, x, p=3, eps=1e-6):\n return F.avg_pool2d(x.clamp(min=eps).pow(p), (x.size(-2), x.size(-1))).pow(1./p)\n \n def __repr__(self):\n return self.__class__.__name__ + \\\n '(' + 'p=' + '{:.4f}'.format(self.p.data.tolist()[0]) + \\\n ', ' + 'eps=' + str(self.eps) + ')'","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:15.412765Z","iopub.execute_input":"2023-10-28T02:38:15.413796Z","iopub.status.idle":"2023-10-28T02:38:15.420086Z","shell.execute_reply.started":"2023-10-28T02:38:15.413752Z","shell.execute_reply":"2023-10-28T02:38:15.418986Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#

Create Model

","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\nclass UBCModel(nn.Module):\n def __init__(self, model_name, num_classes, pretrained=False):\n super(UBCModel, self).__init__()\n self.model = timm.create_model(model_name, pretrained=pretrained)\n if \"efficient\" in model_name:\n in_features = self.model.classifier.in_features\n self.model.classifier = nn.Identity()\n self.model.global_pool = nn.Identity()\n else: # ConvNext\n in_features = self.model.head.in_features\n self.model.head = nn.Identity()\n self.pooling = GeM()\n self.linear = nn.Linear(in_features, num_classes)\n self.softmax = nn.Softmax(dim=1)\n\n def forward(self, images):\n features = self.model(images)\n pooled_features = self.pooling(features).flatten(1)\n output = self.linear(pooled_features)\n return output\n \nmodel = UBCModel(CONFIG['model_name'], CONFIG['num_classes'])\nmodel.load_state_dict(torch.load( BEST_WEIGHT ))\nmodel.to(CONFIG['device']);","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:15.469074Z","iopub.execute_input":"2023-10-28T02:38:15.469868Z","iopub.status.idle":"2023-10-28T02:38:15.475846Z","shell.execute_reply.started":"2023-10-28T02:38:15.469838Z","shell.execute_reply":"2023-10-28T02:38:15.474922Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Prepare Dataloaders","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\ntest_dataset = UBCDataset(df_crop, transforms=data_transforms[\"valid\"])\ntest_loader = DataLoader(test_dataset, batch_size=CONFIG['valid_batch_size'], \n num_workers=2, shuffle=False, pin_memory=True)","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:15.477642Z","iopub.execute_input":"2023-10-28T02:38:15.478051Z","iopub.status.idle":"2023-10-28T02:38:15.488219Z","shell.execute_reply.started":"2023-10-28T02:38:15.47802Z","shell.execute_reply":"2023-10-28T02:38:15.487356Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Start Inference","metadata":{}},{"cell_type":"code","source":"%%writefile -a infer.py\npreds = []\nwith torch.no_grad():\n bar = tqdm(enumerate(test_loader), total=len(test_loader))\n for step, data in bar: \n images = data['image'].to(CONFIG[\"device\"], dtype=torch.float) \n batch_size = images.size(0)\n outputs = model(images)\n outputs = model.softmax(outputs)\n preds.append( outputs.detach().cpu().numpy() )\n\npreds = np.vstack(preds)\nnp.save(\"preds.npy\", preds)\nprint(preds.shape)","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:15.498952Z","iopub.execute_input":"2023-10-28T02:38:15.499234Z","iopub.status.idle":"2023-10-28T02:38:15.504521Z","shell.execute_reply.started":"2023-10-28T02:38:15.499188Z","shell.execute_reply":"2023-10-28T02:38:15.503647Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\nfor i in range(preds.shape[-1]):\n df_crop[f\"cat{i}\"] = preds[:, i]\n\ndict_label = {}\ndict_conf = {}\nfor image_id, gdf in df_crop.groupby(\"image_id\"):\n conf = gdf[ [f\"cat{i}\" for i in range(preds.shape[-1])] ].values.max(axis=0)\n dict_conf[image_id] = np.max(conf)\n dict_label[image_id] = np.argmax( conf )\n #dict_label[image_id] = np.argmax( gdf[ [f\"cat{i}\" for i in range(preds.shape[-1])] ].values.mean(axis=0) )\npreds = np.array( [ dict_label[image_id] for image_id in df[\"image_id\"].values ] )\nconfs = np.array( [ dict_conf[image_id] for image_id in df[\"image_id\"].values ] )","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:15.538745Z","iopub.execute_input":"2023-10-28T02:38:15.539385Z","iopub.status.idle":"2023-10-28T02:38:15.545297Z","shell.execute_reply.started":"2023-10-28T02:38:15.539356Z","shell.execute_reply":"2023-10-28T02:38:15.544403Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%writefile -a infer.py\npred_labels = encoder.inverse_transform( preds )\ndf_sub[\"label\"] = pred_labels\ndf_sub[\"conf\"] = confs\ndf_sub[[\"image_id\", \"label\"]].to_csv(\"submission.csv\", index=False)\ndf_sub.to_csv(\"log.csv\", index=False)","metadata":{"execution":{"iopub.status.busy":"2023-10-28T02:38:15.546877Z","iopub.execute_input":"2023-10-28T02:38:15.547148Z","iopub.status.idle":"2023-10-28T02:38:15.55582Z","shell.execute_reply.started":"2023-10-28T02:38:15.547125Z","shell.execute_reply":"2023-10-28T02:38:15.555006Z"},"trusted":true},"execution_count":null,"outputs":[]}]}