search

Yaoyi-Li / GCA-Matting

Official repository for Natural Image Matting via Guided Contextual Attention
MIT License
390 stars 62 forks source link

Request - Add A Pre-trained model #1

Closed ofirkris closed 4 years ago

ofirkris commented 4 years ago

Hi, thanks for sharing this great research. Can you kindly share a pre-trained model?

Yaoyi-Li commented 4 years ago

Please have a wait. I'm writing the README.md

tianmingdu commented 4 years ago

Is it possible to release the paper 'GCA Matting' online?

Yaoyi-Li commented 4 years ago

Is it possible to release the paper 'GCA Matting' online?

I released this paper on arxiv https://arxiv.org/pdf/2001.04069.pdf and updated the README. 😃

ofirkris commented 4 years ago

Thanks for sharing your work. I'm researching ways to implement image matting on scenarios which require an extremely lightweight model.

Do you think it's possible to achieve a notable reduction in model parameters + model size (current is 101.3MB) while keeping accuracy at high level?

Yaoyi-Li commented 4 years ago

I think it depends on how you define the high level accuracy. I also have tried some extremely lightweight models on mobile devices like https://arxiv.org/pdf/1905.06747.pdf. But in that case, the task is more difficult than image matting with trimap. The predictions of that model are not very stable and highly depends on the quality of input images. IndexNet shows that mobileNet works fine in image matting. So I think it is possible to have a extremely lightweight matting model and we are also working on this.

ofirkris commented 4 years ago

Thanks @Yaoyi-Li - indeed - they also have a web demo here - http://matting.zsyhh.com:4800/ End result with GCA matting is superior also due to the manual trimap needed.