Question about adding mask orders image.

[phamkhactu]

Thanks for great working.

I tried stitch images with order images. As the question in stackoverflow. If I have a mask matrix for Apply2 function, function auto works with orders. With Kornia, how can I pass mask matrix for stitching.

you can provide some code example if you have feature like that, if not, I think it maybe is a feature request for making Koria package more useful.
Thank you very much!

[edgarriba]

Have you tried zeroing the image based on the mask (hadamard prod) ? /cc @ducha-aiki

[phamkhactu]

Have you tried zeroing the image based on the mask

Yes, I have. I run it successfully with order images.

[edgarriba]

@phamkhactu does it mean that it solved your problem ?

[phamkhactu]

@phamkhactu does it mean that it solved your problem ?

yes, it does, it helps to reduce time for matching picture with picture. That is why I require feature.

[edgarriba]

if i understand correctly you are asking for a mask parameter that internally applies elementwise product to the input images ? If that's the case, i'm not quite sure we might go for it since it's a pre-step that the user can always apply.

[phamkhactu]

if i understand correctly you are asking for a mask parameter that internally applies elementwise product to the input images ? If that's the case, i'm not quite sure we might go for it since it's a pre-step that the user can always apply.

Yes, IF I can give mask, maybe algorithm not need calculate relative images. Some time, calculate not right. So can give it in function becomes better.

[edgarriba]

i don't the model architecture is designed for such scenario. Seems more like a research project to fine tune the network (which we don't provide the code) and input RGB+mask to learn more accurate attention maps

[phamkhactu]

i don't the model architecture is designed for such scenario. Seems more like a research project to fine tune the network (which we don't provide the code) and input RGB+mask to learn more accurate attention maps

Yeah,
Thank you for more your details information.