How to convert pred_masks to polygons?

[miladfa7]

In my work, I need to convert the model output of polygon points instead of pred_masks.

In the annotated dataset, an object in an image, we have 15 segmentation points that we do not have in the model output(pred_masks). What can we do to have these segmentation points for objects?

[zhanghang1989]

If the dataset is generated by model using pseudolabel, you can simply set

detectron2/detectron2/config/defaults.py

Line 87 in e3053c1

_C.INPUT.MASK_FORMAT = "polygon" # alternative: "bitmask"

to bitmask.