refactor to avoid NoneType in optimize_for_inference

[isaacrob-roboflow]

Description

Fixes an issue where during JIT tracing of an object detector a NoneType would be returned for a placeholder for the mask object which is incompatible with JIT. Fixed by returning a variable-length tuple instead

Type of change

Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)

How has this change been tested, please provide a testcase or example of how you tested the change?

tested locally

Any specific deployment considerations

For example, documentation changes, usability, usage/costs, secrets, etc.

Docs

• Docs updated? What were the changes:

[software-ai-life]

Good catch overall 👍
However, there's a small logical issue here:

After converting predictions from a tuple to a dict,
the length check (len(predictions) == 3) will never be true,
so pred_masks will never be added to the output.

Suggested fix:

if len(predictions) == 3:
    predictions = {
        "pred_boxes": predictions[0],
        "pred_logits": predictions[1],
        "pred_masks": predictions[2],
    }
else:
    predictions = {
        "pred_boxes": predictions[0],
        "pred_logits": predictions[1],
    }