Can you please create new CoreML model for this BeautyGan?

[ioskevinshah]

https://github.com/Honlan/BeautyGAN

[john-rocky]

I tried but failed. 2022年4月1日(金) 21:13 ioskevinshah ***@***.***>:

…

https://github.com/Honlan/BeautyGAN — Reply to this email directly, view it on GitHub <#14>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFRTLEF66JPKV3UDRGGAGILVC3R6DANCNFSM5SIQHZPA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[ioskevinshah]

is this possible to do in iOS?

https://rockyshikoku.medium.com/refine-rough-segmentation-a24c9bdf1b42

[john-rocky]

also failed🥺

[john-rocky]

But I want to convert those, so I will revenge .

[ioskevinshah]

(https://developer.apple.com/documentation/arkit/armattegenerator)

Apple is providing matte generator, but this limited to ARKit. also, I tired but not accurate.

[ioskevinshah]

How about this?

https://github.com/TencentARC/GFPGAN

to CoreML

[john-rocky]

I tried, but color taste of output was bad.
I think I missed toRGB ops.

[john-rocky]

For segment people, is it not enough to use person segmentation of Vision?

[ioskevinshah]

Also, please check style transfer link for Colab.

https://colab.research.google.com/drive/1f6KOh9fuOmCP9vbNldqmyX6lqReGVtr4?usp=sharing

[ioskevinshah]

For segment people, is it not enough to use person segmentation of Vision?
-> Yes, I've also modified the u2net to 512*512 because the edges were not smooth in 320 by your model.

That's why I ask for refine rough segmentation. Apple iOS 15 vision fair enough, but is limited to person only.

[john-rocky]

Sorry, I lost conversion script.
Please refer this.

https://rockyshikoku.medium.com/coreml-conversion-that-enables-fast-neural-style-transfer-to-be-used-on-ios-and-macos-d66f02a2489f

Thank you for let me know.

[ioskevinshah]

Thanks, you are doing amazing work. Sooner, I'll notified with my own app. Just final testing is going on. I think you will going to love this app.

[ioskevinshah]

Also, please check out.
#10 (comment)

[ioskevinshah]

Hello @john-rocky ,

Can please provide, UGATIT on 512 and 768 input image size models?.

Thanks,
Kevin Shah