search

YutingXiao / Amodal-Segmentation-Based-on-Visible-Region-Segmentation-and-Shape-Prior

Implementation of the AAAI-2021 paper "Amodal Segmentation Based on Visible Region Segmentation and Shape Prior"
52 stars 7 forks source link

How did you get 80 classes for COCO-A dataset? #16

Open zbwxp opened 2 years ago

zbwxp commented 2 years ago

Hi @YutingXiao I really appreciate this nice job. We are planning to present another work on amodal segmentation and want to follow this work you have done.

In your paper, you said the COCO-A dataset has 80 classes. I can no longer download the annotations you provided from the link

_The COCOA dataset annotation: ftp://guest:GU.205dldo@ftp.softronics.ch/cocoa/COCOA_annotationsdetectron.tar.xz

Since it kept saying "the password is incorrect" Thus, I got the original COCOA dataset annotation. However, the raw annotation has a "name" in every region together having 2140 different descriptions. How did you transfer them into 80 classes?

I saw in your detectron2/data/datasets/builtin.py that it has a "coco_2014_amodal_train" and a "cocoa_cls_train" so I assume this new "cocoa_cls_train" is generated by yourself.

Could you please provide the rule for mapping 2140 names to 80 classes? Or it would be a great help if you can provide your original amodal_cls_annotions.

Thank you very much.