Post processing process issues

[diya-he]

Hello, while reading the Yolact++ code, I found that the post-processing process first went through FastNMS, and then filtered the mask_iou*score scores for masks and boxes. Is my understanding correct? If so, I hope you can tell me why you are doing this and the significance of doing so. The code is as follows:

output_utils.py:

And for this, we use scores as: scores = [scores, scores * maskiou_p]
detection.py:

We use fast_nms.

and in postprocess, I found you sort by scores and second screening