ValueError(f"Expected {name} for bbox {bbox} to be in the range [0.0, 1.0], got {value}.")

[alejopaullier96]

🐛 Bug

I am getting the following bug which was already addressed here but apparently the bug still persists in albumentations version 1.3.0:

ValueError(f"Expected {name} for bbox {bbox} to be in the range [0.0, 1.0], got {value}.")
ValueError: Expected y_max for bbox (0.14792899408284024, 0.9671641791044776, 0.28205128205128205, 1.0029850746268656, tensor(2)) to be in the range [0.0, 1.0], got 1.0029850746268656.

I am using the following augmentations for an object detection problem:

def get_train_transforms():
    return A.Compose(
        [
            A.Resize(height=config.RESOLUTION, width=config.RESOLUTION, p=1),
            A.Normalize(p=1),
            ToTensorV2(p=1.0),
        ], 
        p=1.0, 
        bbox_params=A.BboxParams(
            format='pascal_voc',
            min_area=0, 
            min_visibility=0,
            label_fields=['labels']
        )
    )

def get_valid_transforms():
    return A.Compose(
        [
            A.Resize(height=config.RESOLUTION, width=config.RESOLUTION, p=1.0),
            A.Normalize(p=1),
            ToTensorV2(p=1.0),
        ], 
        p=1.0, 
        bbox_params=A.BboxParams(
            format='pascal_voc',
            min_area=0, 
            min_visibility=0,
            label_fields=['labels']
        )
    )

Expected behavior

Bounding boxes should not be larger than image dimensions after transformations.

Environment

• Albumentations version: 1.3.0
• Python version (e.g., 3.7): 3.7.12
• OS (e.g., Linux): Linux-5.15.89+-x86_64-with-debian-bullseye-sid
• How you installed albumentations (conda, pip, source): pip
• Any other relevant information: Run with GPU on a Kaggle environment (2023-03-29)

Additional context

GPU info: Cuda: Tesla P100-PCIE-16GB

[alejopaullier96]

After much debugging the problem was due to negative values in bounding boxes data for their x and y coordinates as well as heights and widths of bounding boxes.

[AriyaRasekh]

The bug is defiantly there, I'm assuming some numbers after decimal point somehow is messing up the calculation in the backed, you can try limiting the numbers (your bbox label values) after decimal point to 6 or 10 and try again.

Library has critical bugs for sure. You might have to find a new library for augmentation if the issue persists for you

[ternaus]

@alejopaullier96 @AriyaRasekh

The issue could be resolved if one passes clip=True to BboxParams