Python get_trainval_datasets示例

示例#1

文件： validate_baseline_lwrefinenet.py 项目： vfdev-5/UNOSAT_Challenge

num_classes = 2

#################### Dataflow ####################

assert "INPUT_PATH" in os.environ
data_path = os.path.join(os.environ['INPUT_PATH'], "train_tiles")
csv_path = os.path.join(data_path, "tile_stats.csv")

train_folds = [0, 1, 3]
val_folds = [
    2,
]

train_ds, val_ds = get_trainval_datasets(data_path,
                                         csv_path,
                                         train_folds=train_folds,
                                         val_folds=val_folds)

batch_size = 32
num_workers = 12

mean = (0.0, 0.0, 0.0)
std = (5.0, 5.0, 5.0)
max_value = 1.0

transforms = A.Compose(
    [A.Normalize(mean=mean, std=std, max_pixel_value=max_value),
     ToTensorV2()])

_, data_loader, _ = get_train_val_loaders(
    train_ds,

示例#2

start_by_validation = False

#################### Dataflow ####################

assert "INPUT_PATH" in os.environ
data_path = os.path.join(os.environ['INPUT_PATH'], "train_tiles")
csv_path = os.path.join(data_path, "tile_stats.csv")

train_folds = [0, 1, 3]
val_folds = [
    2,
]

train_ds, val_ds = get_trainval_datasets(
    data_path,
    csv_path,
    train_folds=train_folds,
    val_folds=val_folds,
    read_img_mask_fn=read_img_5b_in_db_with_mask)

train_sampler = get_train_sampler(train_ds, weight_per_class=(0.5, 0.5))
# ! This wont work in distributed !
# mean, std = get_train_mean_std(train_ds, unique_id="3b_in_db")
# print("Computed mean/std: {} / {}".format(mean, std))
mean = [
    -17.704988005545587, -10.33310725243658, -12.422949109368183,
    213.3866453581477, 0.4748089840110086
]
std = [
    6.5437130712772795, 6.033536195001276, 6.063934363438651,
    245.40096009414592, 238.8577452846451
]

示例#3

文件： baseline_lwrefinenet_sqrt.py 项目： vfdev-5/UNOSAT_Challenge

start_by_validation = False

#################### Dataflow ####################

assert "INPUT_PATH" in os.environ
data_path = os.path.join(os.environ['INPUT_PATH'], "train_tiles")
csv_path = os.path.join(data_path, "tile_stats.csv")

train_folds = [0, 1, 3]
val_folds = [
    2,
]

train_ds, val_ds = get_trainval_datasets(data_path,
                                         csv_path,
                                         train_folds=train_folds,
                                         val_folds=val_folds,
                                         read_img_mask_fn=read_nimg_sqrt_mask)

train_sampler = get_train_sampler(train_ds, weight_per_class=(0.7, 0.3))

batch_size = 22
num_workers = 12
val_batch_size = 20

# According to https://arxiv.org/pdf/1906.06423.pdf
# For example: Train size: 224 -> Test size: 320 = max accuracy on ImageNet with ResNet-50
val_img_size = 512
train_img_size = 480

train_transforms = A.Compose([