|
from ..data_aug import cifar_like_image_train_aug, cifar_like_image_test_aug, imagenet_like_image_test_aug, imagenet_like_image_train_aug |
|
from ..ab_dataset import ABDataset |
|
from ..dataset_split import train_val_test_split |
|
from torchvision.datasets import ImageFolder |
|
from typing import Dict, List, Optional |
|
from torchvision.transforms import Compose |
|
from utils.common.others import HiddenPrints |
|
from ..dataset_cache import get_dataset_cache_path, read_cached_dataset_status, cache_dataset_status |
|
import os |
|
from ..registery import dataset_register |
|
|
|
|
|
@dataset_register( |
|
name='SYNSIGNS', |
|
classes=[f'{i:05d}' for i in range(42)], |
|
task_type='Image Classification', |
|
object_type='Traffic Sign', |
|
class_aliases=[], |
|
shift_type=None |
|
) |
|
class SYNSIGNS(ABDataset): |
|
def create_dataset(self, root_dir: str, split: str, transform: Optional[Compose], |
|
classes: List[str], ignore_classes: List[str], idx_map: Optional[Dict[int, int]]): |
|
if transform is None: |
|
transform = imagenet_like_image_train_aug() if split == 'train' else imagenet_like_image_test_aug() |
|
self.transform = transform |
|
|
|
dataset = ImageFolder(root_dir, transform=transform) |
|
|
|
cache_file_path = get_dataset_cache_path(root_dir, classes, ignore_classes, idx_map) |
|
if os.path.exists(cache_file_path): |
|
dataset.samples = read_cached_dataset_status(cache_file_path, 'SYNSIGNS-' + split) |
|
else: |
|
if len(ignore_classes) > 0: |
|
ignore_classes_idx = [classes.index(c) for c in ignore_classes] |
|
dataset.samples = [s for s in dataset.samples if s[1] not in ignore_classes_idx] |
|
if idx_map is not None: |
|
dataset.samples = [(s[0], idx_map[s[1]]) if s[1] in idx_map.keys() else s for s in dataset.samples] |
|
cache_dataset_status(dataset.samples, cache_file_path, 'SYNSIGNS-' + split) |
|
|
|
dataset = train_val_test_split(dataset, split) |
|
return dataset |
|
|
