Random_resize_crop() for ImageBBoxes

Hi, I’m working on object detection and want to add further data augmentations, like rand_resize_crop().
The following code results in an error (note: I only applied the first transform function from rand_resize_crop() to find the error, hence the [0]):

t = torch.eye(300, 400).unsqueeze(0)
image = Image(t)

box = torch.Tensor([0.25, 0.25, 0.75, 0.75]).unsqueeze(0) * torch.Tensor([*image.size, *image.size])
box_ibb = ImageBBox.create(*image.size, box)

tfms = [rand_resize_crop(size=300)[0]]
image_t = image.apply_tfms(tfms, do_resolve=True)
box_t = box_ibb.apply_tfms(tfms, do_resolve=False)

The causing error in the call stack is this:

~\Documents\Python Scripts\fastai\vision\transform.py in _compute_zs_mat(sz, scale, squish, invert, row_pct, col_pct)
310 “Utility routine to compute zoom/squish matrix.”
311 orig_ratio = math.sqrt(sz[1]/sz[0])
–> 312 for s,r,i in zip(scale,squish, invert):
313 s,r = 1/math.sqrt(s),math.sqrt®
314 if s * r <= 1 and s / r <= 1: #Test if we are completely inside the picture

TypeError: zip argument #3 must support iteration

Which is caused by the fact that it tries to iterate on invert, which is just True.
While scale and squish are tensors. Following the code, the cause for invert being a single boolean is (in image.py:290):

def coord(self, func:CoordFunc, *args, **kwargs)->'ImagePoints':
    "Put `func` with `args` and `kwargs` in `self.flow_func` for later."

    if 'invert' in kwargs: kwargs['invert'] = True
    else: warn(f"{func.__name__} isn't implemented for {self.__class__}.")
    self.flow_func.append(partial(func, *args, **kwargs))
    return self

If I remove the if-else-block (which might be there for a reason) the code runs and transforms the ImageBBox.
However the resulting transformation is not correct, which can be seen in the image:

The top image is the image and bbox before transformation and below after transform. Right now I’m a bit stucked to further follow the chain to find the reason for that.
Maybe someone of you can help find the cause.
(I’m using fast.ai 1.0.46)