Open peko opened 3 years ago
Hi, Vladimir,
Thanks a lot for your interests. The results look pretty good, especially the videos. We thought about producing some video samples, but don't have time to do that yet. Can I show one of your video results in this repo ? Of course, this credit belongs to you.
Regards, Xuebin
On Sat, Nov 28, 2020 at 5:55 PM Vladimir Seregin notifications@github.com wrote:
I want to say that the result is amazing and inspiring. The guys from the Disney laboratory have been struggling for years over the tasks of non-photo-realistic render and line art stylization, they write heaps of articles about non-solvable problems in this direction.
The main problem is that this style contains a lot of symbolism that is contrary to the physics of lighting. This is especially true for parts of the face - eyes, lips, face contour, hair. It is just too hard to catch such nuances, genre traditions, and this model almost do it.
A few years ago I was developing an ink drawing robot and the main problem was getting a high-quality contour for the brush trajectory. And your approach greatly simplifies the task.
I have uploaded my tests https://yadi.sk/d/Df7ecM3LtXcNOA?w=1, dir contains some thousands processed portrait photos including SR version and video test. Original photos can be found at readme file.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/NathanUA/U-2-Net/issues/106, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORIB3T2IZOAMTGJVHP3SSGLYZANCNFSM4UGEV5QQ .
-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/
Of course, it's up to you. I do not know if you can use the results of processing random videos from YouTube. I just picked up the first one I found with a soft light and a clear focus.
I left a link to the original video in the readme. If you have any problems with the source, I will repeat the processing on any other video that suits you.
Hi, I would like to share with you one small tool that I use when checking the quality of the models. It allows you to see the entire photo-dataset or results and quickly find problematic areas.
But in this case I just hard-code compassion between u2net and artline, it just allows you to quickly switch between two or more models and observe the differences (keys 2, 3).
That looks great, Vladimir. I also included your tool in the ReadMe file of U^2-Net so that others can find and use that very easily.
On Tue, Dec 29, 2020 at 4:02 AM Vladimir Seregin notifications@github.com wrote:
Hi, I would like to share with you one small tool https://github.com/peko/nn-lineart that I use when checking the quality of the models. It allows you to see the entire photo-dataset or results and quickly find problematic areas.
But in this case I just hard-code compassion between u2net and artline, it just allows you to quickly switch between two different models and observe the difference (keys 2, 3).
Interactive version here https://peko.github.io/nn-lineart/
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NathanUA/U-2-Net/issues/106#issuecomment-752036084, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORJNJIRTXW52CQOXJTDSXGZKZANCNFSM4UGEV5QQ .
-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/
BTW, Vladimir. Would mind that I include the following gif of your NN based lineart in the ReadMe file of U^2-Net as well. I think that will attract more attention to both of our repos. [image: image.png]
On Tue, Dec 29, 2020 at 7:51 PM Xuebin Qin xuebin@ualberta.ca wrote:
That looks great, Vladimir. I also included your tool in the ReadMe file of U^2-Net so that others can find and use that very easily.
On Tue, Dec 29, 2020 at 4:02 AM Vladimir Seregin notifications@github.com wrote:
Hi, I would like to share with you one small tool https://github.com/peko/nn-lineart that I use when checking the quality of the models. It allows you to see the entire photo-dataset or results and quickly find problematic areas.
But in this case I just hard-code compassion between u2net and artline, it just allows you to quickly switch between two different models and observe the difference (keys 2, 3).
Interactive version here https://peko.github.io/nn-lineart/
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NathanUA/U-2-Net/issues/106#issuecomment-752036084, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORJNJIRTXW52CQOXJTDSXGZKZANCNFSM4UGEV5QQ .
-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/
-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/
I want to say that the result is amazing and inspiring. The guys from the Disney laboratory have been struggling for years over the tasks of non-photo-realistic render and line art stylization, they write heaps of articles about non-solvable problems in this direction.
The main problem is that this style contains a lot of symbolism that is contrary to the physics of lighting. This is especially true for parts of the face - eyes, lips, face contour, hair. It is just too hard to catch such nuances, genre traditions, and this model almost do it.
A few years ago I was developing an ink drawing robot and the main problem was getting a high-quality contour for the brush trajectory. And your approach greatly simplifies the task.
I have uploaded my tests, dir contains some thousands processed portrait photos including SR version and video test. Original photos can be found at readme file.