torch script c++ mask rcnn inference output

[abucka]

Hello,

while running a mask rcnn inference, I get pred_masks output as implemented in tools/deploy/torchscript_mask_rcnn.cpp. the pred_masks have a size (3,80,28,28), I understand that 3 is the number of instances and (28,28) is the size of masks that should be resized to the input image size. what does the dimension 80 means in this case? how can I perform postprocessing in order to visualise the masks?