Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

How to improve accuracy? #11

Open
InternetMaster1 opened this issue Mar 15, 2020 · 4 comments
Open

How to improve accuracy? #11

InternetMaster1 opened this issue Mar 15, 2020 · 4 comments

Comments

@InternetMaster1
Copy link

I tried the demo on the android app using the provided pre trained models. But the cutout is not accurate

How to increase the accuracy of the mask?

Is it possible to train a lot more and get results like remove.bg?

@nolanliou
Copy link
Owner

  1. Use better trainning dataset.
  2. Add post-processing module.
  3. Use better backbone

@InternetMaster1
Copy link
Author

Thank you so much for the answer. I am new to machine learning. Your help is much appreciated.

  1. I was planning to use the aisegment dataset and the Supervisely person. This should be enough? For how many steps should we train to get good results?

  2. Could you please explain about post processing module or provide some link?

  3. Any suggestion on what would be a better backbone?

Thanks for being patient with this newbie

@InternetMaster1
Copy link
Author

InternetMaster1 commented May 2, 2020

Hi,

  1. Thanks

  2. Thanks

  3. Backbone :- Please help!

Would resnet50 be better for accuracy? or portraitnet, etc?
Could you point me towards some evaluation metrics or a chart where there might be comparison of various backbones?

Any guidance on the backbone would be most appreciated! Thanks

@mapleyuan
Copy link

@InternetMaster1 hi, could you shared your datasets or project? I face the same problem. thanks!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants