search

xuebinqin / DIS

This is the repo for our new project Highly Accurate Dichotomous Image Segmentation
Apache License 2.0
2.21k stars 256 forks source link

About the preparation of the DIS dataset and the next research topic. #23

Closed faruknane closed 2 years ago

faruknane commented 2 years ago

@xuebinqin Hii, do you have any chance to tell a bit about how you prepared and annotated the data V1 and V2?

For the last 10 days, I have been annotating some high res data for my validation dataset. I got 66 images annotated carefully. I worked on pre-predicted masks and fixed them. Thus, the process was easier for me. Although I did a little work, the burden was huge.

So, how did you manage to generate or annotate the data? It doesn't look like any artificial (or rendered) data to me.

Also, do you have any more ongoing research on image segmentation topic? What's next?

Thank you!

xuebinqin commented 2 years ago

Hi, Akif,

Thanks for your interest. Yes, all the images are real natural images annotated by us manually. That is one of the reasons why it takes us more than a year for this paper. We do summarize some of the skills for manual annotation. But the workload is still huge. For DIS V1.0, the average labeling time for each image is 0.5 hours and some images take us up to 10 hours. In DIS V2.0 (unreleased one), the images' complexities are more diversified and some images take up to 70 hours (the peacock in our github) for labeling. We are trying to develop some semi-automatic ways and self-supervised ways for segmenting highly accurate results. We probably prepare a tutorial for annotating these highly accurate masks and share it with the whole community later.

On Tue, Aug 2, 2022 at 3:58 AM Akif Faruk Nane @.***> wrote:

@xuebinqin https://github.com/xuebinqin Hii, do you have any chance to tell a bit about how you prepared and annotated the data V1 and V2?

For the last 10 days, I have been annotating some high res data for my validation dataset. I got 66 images annotated carefully. I worked on pre-predicted masks and fixed them. Thus, the process was easier for me. Although I did a little work, the burden was huge.

So, how did you manage to generate or annotate the data? It doesn't look like any artificial (or rendered) data to me.

Also, do you have any more ongoing research on image segmentation topic? What's next?

Thank you!

— Reply to this email directly, view it on GitHub https://github.com/xuebinqin/DIS/issues/23, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORKVF4LQV2R7N42L7XTVXD5MLANCNFSM55KWOV2Q . You are receiving this because you were mentioned.Message ID: @.***>

-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage: https://xuebinqin.github.io/

faruknane commented 2 years ago

@xuebinqin honestly, great work both in developing a new type of AI model (IS-NET) and also annotate all te data (DIS5K)! I wonder your data annotation methods. We appreciate you offering these to the community. Thank you much for your answer!