I tried running my inference with the runner.run after setting _`__call__`_ in the signatures. It threw an exception:

TypeError: 'Detections' object is not subscriptable

It seems like it happens at the batching mechanism when handling the results of the inference