A.Oneof报概率分母为 0 的错误

[Charon-yzc]

A.OneOf([
IsotropicResize(...),
IsotropicResize(...),
IsotropicResize(...),
], p = 0 if self.config['with_landmark'] else 1)

Traceback (most recent call last): File "/root/autodl-tmp/DeepfakeBench/training/train.py", line 323, in main() File "/root/autodl-tmp/DeepfakeBench/training/train.py", line 278, in main train_data_loader = prepare_training_data(config) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/root/autodl-tmp/DeepfakeBench/training/train.py", line 87, in prepare_training_data train_set = DeepfakeAbstractBaseDataset( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/root/autodl-tmp/DeepfakeBench/training/dataset/abstract_dataset.py", line 111, in init self.transform = self.init_data_aug_method() ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/root/autodl-tmp/DeepfakeBench/training/dataset/abstract_dataset.py", line 141, in init_data_aug_method A.OneOf([ File "/root/miniconda3/lib/python3.12/site-packages/albumentations/core/composition.py", line 987, in init self.transforms_ps = [t / s for t in transforms_ps] ^ ZeroDivisionError: float division by zero

出现ZeroDivisionError: float division by zero

[KaguraNaku]

You can fix this by modifying dataset/albu.py to use:
super(IsotropicResize, self).__init__(p=1 if always_apply else p) instead of the original super call super(IsotropicResize, self).__init__(always_apply, p).
This issue occurs in Albumentations v2.x (verified in v2.0.8). In the 2.x release, the always_apply parameter was deprecated from the base class __init__.
In the buggy custom class IsotropicResize, always_apply=False is passed as the first positional argument. Since the first argument is now p, the boolean False is automatically cast to the float 0.0, causing a ZeroDivisionError to occur.
You can also see this explained at Albumentations 2.0.0 Release Notes

Deleted always_apply => use p=1 to always apply and p=0 for not applying.