search

xuebinqin / U-2-Net

The code for our newly accepted paper in Pattern Recognition 2020: "U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection."
Apache License 2.0
8.67k stars 1.5k forks source link

background segmentation #223

Open xiongzhu666 opened 3 years ago

xiongzhu666 commented 3 years ago

Hi~I retrain a sky segmentation based on your project,I have some doubt：

  1. my dataset is sky mask，the mask is alpha mask and then I binary the mask(0/255). so that the binary mask is not very precise or not too bad（for example the tree branches.....），I don't know if it will affect the training result.
  2. base on my doubt.1, do you think it will improve by use L1 loss function?? I mean use alpha mask not binary mask about training labels

thanks best regards

bluesky314 commented 3 years ago

Can you share your data @xiongzhu666 ? Would love to work on sky segmentation

xiongzhu666 commented 3 years ago

@xuebinqin thanks for your great working....I get a good result based on your repository for sky matting 2344235

bluesky314 commented 3 years ago

@xiongzhu666 Which sky matting data are you using?

xuebinqin commented 3 years ago

Hi, Xiongzhu,

Thanks for your interest. The results look awesome.

On Fri, Jul 2, 2021 at 6:34 AM xiongzhu666 @.***> wrote:

@xuebinqin https://github.com/xuebinqin thanks for your great working....I get a good result based on your repository for sky matting [image: 2344235] https://user-images.githubusercontent.com/52955431/124212100-eb6a9800-db20-11eb-928b-76e49fd1ee28.png

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/xuebinqin/U-2-Net/issues/223#issuecomment-872668775, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORLTYMJWVAG2CCKJVRTTVUQSBANCNFSM47OBYEKA .

-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

xiongzhu666 commented 3 years ago

@xiongzhu666 Which sky matting data are you using?

@bluesky314 sorry to reply late, I just use about 9000 ADE20K datasets and 5000 company's private data. please refer my result: https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing