How do I make my own dataset mask？Looking forward to your reply

[tamnguyenvan]

Same question

[SwayamInSync]

@xuebinqin can you please help with this issue, I also need to create masks and DIS5K dataset's masks are too accurate.

[xuebinqin]

Sorry, what do you mean about too accurate? You can get your jpg image as model input and prepare your ground truth mask as binarized image with either pixel value of 0 or 255. The ground truth can be created by gimp: https://www.gimp.org/.

On Tue, Jul 26, 2022 at 8:58 AM Swayam @.***> wrote:

@xuebinqin https://github.com/xuebinqin can you please help with this issue, I also need to create masks and DIS5K dataset's masks are too accurate.

— Reply to this email directly, view it on GitHub https://github.com/xuebinqin/DIS/issues/16#issuecomment-1195669526, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGOROIKO3W5H2IVKBJ663VWADKXANCNFSM54QXBVAQ . 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/

[tamnguyenvan]

@xuebinqin Could you kindly share a short video tutorial on how you used gimp to make the ground truth mask?

[xuebinqin]

please just follow the comments in the https://github.com/xuebinqin/DIS/blob/main/IS-Net/train_valid_inference_main.py

in addition, you need to comment the 294 line out and uncomment the 295 line in " https://github.com/xuebinqin/DIS/blob/main/IS-Net/data_loader_cache.py". It is a temporary solution. we will update the code soon for solving that issue.

On Sun, Jul 24, 2022 at 8:16 PM xiaosi @.***> wrote:

— Reply to this email directly, view it on GitHub https://github.com/xuebinqin/DIS/issues/16, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORM4TVHW7FQJETEZZUTVVYBHHANCNFSM54QXBVAQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

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

[zhanghongyong123456]

Sorry, what do you mean about too accurate? You can get your jpg image as model input and prepare your ground truth mask as binarized image with either pixel value of 0 or 255. The ground truth can be created by gimp: https://www.gimp.org/. … On Tue, Jul 26, 2022 at 8:58 AM Swayam @.> wrote: @xuebinqin https://github.com/xuebinqin can you please help with this issue, I also need to create masks and DIS5K dataset's masks are too accurate. — Reply to this email directly, view it on GitHub <#16 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGOROIKO3W5H2IVKBJ663VWADKXANCNFSM54QXBVAQ . 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/

I think he was to say: the masks of the datasets you released are so good（mean accuracy），It's hard to imagine this was done by hand；Very impressive！

[xuebinqin]

Thanks for your suggestion. We will think about it and probably provide a simple tutorial for the labeling skills, which are sometimes tricky but valuable to most of the users.

On Tue, Jul 26, 2022 at 8:01 PM Nguyễn Văn Tâm @.***> wrote:

@xuebinqin https://github.com/xuebinqin Could you kindly share a short video tutorial on how you used gimp to make the ground truth mask?

— Reply to this email directly, view it on GitHub https://github.com/xuebinqin/DIS/issues/16#issuecomment-1196211760, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORJUPW6YLHG4VUM57N3VWCRCBANCNFSM54QXBVAQ . 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/

[kabbas570]

Sorry, what do you mean about too accurate? You can get your jpg image as model input and prepare your ground truth mask as binarized image with either pixel value of 0 or 255. The ground truth can be created by gimp: https://www.gimp.org/. … On Tue, Jul 26, 2022 at 8:58 AM Swayam @.> wrote: @xuebinqin https://github.com/xuebinqin can you please help with this issue, I also need to create masks and DIS5K dataset's masks are too accurate. — Reply to this email directly, view it on GitHub <#16 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGOROIKO3W5H2IVKBJ663VWADKXANCNFSM54QXBVAQ . 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/

Please mention how to do it for multi-class segmentation. (binarized image with either pixel value of 0 or 255. ) I think this is true for Binary segmentation. Thanks