Missing key error - Inference function does not return image features

In the [mmt release notebook](https://github.com/deepmind/multimodal_transformers/blob/main/mmt_release.ipynb), in the function definition `RunInferenceSingleImage(image, graph)`, nowhere does it return `detection features`.

Yet, after that function call, `output_dict` is instantiated as `output_dict = RunInferenceSingleImage(image_np, detection_graph)` and it tries to access `output_dict['detection_features']`.

Obviously this leads to errors about missing keys.

Any hints to how to resolve this will be appreciated. I think I might be missing something simple.