How is the pre-processing of ms-coco dataset done?

[ptirupat]

Hi, I see that in your code (image_seg_data_layer.cpp) you are reading the segmentation maps (.png) for the ms-coco dataset from a location (root_folder + "SegmentationClass/" + lines_[lines_id_].second + ".png"). I would like to know how did you handle the issue of overlapping segmentations in the ms-coco dataset while generating these segmentation maps.