Segmentation Mask Scores

I want to get a score for each predicted mask in each frame. I found in the output of predictor.handle_stream_request, the field "out_probs" stays constant across frames. Is "out_probs" just the detection score for the first frame?
```
for response in predictor.handle_stream_request(request=dict(type="propagate_in_video", session_id=session_id)):
outputs = response["outputs"]
obj_score_logits = outputs["out_probs"] # stays the same across frames
...
```
Any suggestion of how to get the per-mask scores is appreciated.

Start at predictor.handle_stream_request and the propagate_in_video response shown in the issue. Trace how outputs["out_probs"] is produced and whether it changes across frames, then inspect the surrounding inference outputs for a per-mask score. Done means clearly determining what out_probs represents and documenting or exposing the per-mask scores requested by the reporter.

Written by the indexing model from the issue text.