|
|
| import torch.utils.data as data |
| import os |
| from PIL import Image |
| import torch |
| import logging |
| import numpy as np |
| from .data_path import get_dataset_path |
| from sklearn.model_selection import train_test_split |
|
|
| VINDR_CLASSES = ['Lung Opacity', 'Cardiomegaly', 'Pleural thickening', 'Aortic enlargement', 'Pulmonary fibrosis', 'Tuberculosis',] |
|
|
|
|
| VIN_DEFAULT_PATH = get_dataset_path('vinbigdata') |
|
|
| class VinDrNew(data.Dataset): |
| def __init__(self, root=VIN_DEFAULT_PATH, transform=None, data_pct=1.0, split='train', data_postfix='', return_label=True, dataset_seed=42): |
| super().__init__() |
| root = root + data_postfix |
| logging.info(f'VinDr from {root}') |
| assert split in ('train', 'val', 'test') |
| self.classes = None |
| self.num_classes = 6 |
| dataset_split = 'test' if split == 'test' else 'train' |
| with open(f'annotations/vindr/VinDrCXR_{dataset_split}_pe_global_one.txt', 'r') as f: |
| lines = f.readlines() |
| self.img_list = [] |
| self.labels = [] |
| for l in lines: |
| words = l.strip().split(' ') |
| words = [w for w in words if w != ''] |
| self.img_list.append(os.path.join(root, 'images', words[0]+'.jpg')) |
| self.labels.append([int(l) for l in words[1:]]) |
| |
| if split != 'test': |
| train_img_list, val_img_list, train_labels, val_labels = train_test_split(self.img_list, self.labels, train_size=0.9, random_state=dataset_seed) |
| logging.info(f'Dataset Seed: {dataset_seed}') |
| if split == 'train': |
| self.img_list, self.labels = train_img_list, train_labels |
| else: |
| self.img_list, self.labels = val_img_list, val_labels |
| |
| if data_pct < 1.0 and split == 'train': |
| state = np.random.RandomState(seed=dataset_seed) |
| num_data = len(self.img_list) |
| indices = state.choice(np.arange(num_data), size=int(data_pct * num_data), replace=False) |
| self.img_list = [self.img_list[idx] for idx in indices] |
| self.labels = [self.labels[idx] for idx in indices] |
| self.labels = np.asarray(self.labels).astype(np.float32) |
| logging.info(f'VinDr New {split} split: {len(self.img_list)} images') |
| self.transform = transform |
| self.return_label = return_label |
|
|
| def __len__(self): |
| return len(self.img_list) |
|
|
| def __getitem__(self, index): |
| img_path, label = self.img_list[index], self.labels[index] |
|
|
| img = Image.open(img_path).convert('RGB') |
| if self.transform != None: |
| img = self.transform(img) |
|
|
| if not self.return_label: |
| return img |
| else: |
| return img, torch.FloatTensor(label) |
