Question about box prediction

[OctoNinja9]

In this line (below), it seems that the code uses masks to predict boxes,

X-Decoder/modeling/architectures/xdecoder_model.py

Line 922 in 165f8a6

result.pred_boxes = BitMasks(mask_pred > 0).get_bounding_boxes()

but in line 913, the predicted boxes are already obtained.

X-Decoder/modeling/architectures/xdecoder_model.py

Line 913 in 165f8a6

box_pred = box_pred[keep]

Why do not use predicted bboxes directly?