Detection transforms wrong? Image Transformations Cropping/Resizing Affecting Bounding Box Coordinates

[IvanVassi]

Hi!

I'm currently using monai for object detection and have encountered an issue with image transformations that seem to be affecting the bounding box coordinates. Specifically, when applying resizing and cropping transformations to the images, the corresponding bounding boxes are not transformed correctly, leading to inaccuracies in the detection results.
I've got images in numpy arrays.

train_transforms = Compose(
        [
            LoadImaged(keys=[image_key],reader = 'NumpyReader'),
            Transposed(keys=[image_key], indices = (0,2,3,1)),
            EnsureTyped(keys=[image_key, box_key], dtype=torch.float32),
            EnsureTyped(keys=[label_key], dtype=torch.long),
            EnsureTyped(keys=[image_key], dtype=torch.float16),
            ConvertBoxToStandardModed(box_keys=[box_key], mode="xxyyzz"),
            StandardizeEmptyBoxd(box_keys=["box"], box_ref_image_keys="image"),
            RandCropBoxByPosNegLabeld(
                image_keys=[image_key],
                box_keys=[box_key],
                label_keys=[label_key],
                spatial_size=[192,192,80],
                whole_box=True,
                num_samples=4,
                pos=1,
                neg=1,
            ),        
        ]
    )

Any ideas?