there are too many keys\ids in the object_detection dataset. so error message "image_id" is confusing

for a some reason, there are several keys in the object detection record

• filename
• source_id
• key

but in the code, it's even more confusing - the same field named "hash", "key", "image_id"
so when user creates invalid eval dataset, there are warnings like

"Missing ground-truth for image id: ",
"Ignoring detection with image id %s since it was previously added"

but image_id in the message is nothing from the user-defined keys.
see
https://github.com/tensorflow/models/blob/eb355ec0801fe59600c53e2302c3e1b81ba7e9c0/research/object_detection/inputs.py#L625

and

https://github.com/tensorflow/models/blob/eb355ec0801fe59600c53e2302c3e1b81ba7e9c0/research/object_detection/model_lib.py#L536

it is a hash of the source_id! o.O

Very confusing, and it makes debug dataset issues hard.