-
Notifications
You must be signed in to change notification settings - Fork 103
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
Incomplete coloring with larger --mindim values. #31
Comments
Here are sample videos of the problem - same source and same reference pictures: https://youtu.be/Qfq6L0l4dk8 --mindim 540 https://youtu.be/OjyTnmFb_PY --mindim 320 The red color is the always the first to disappear and fade. |
@kepppis Have you managed to solve this? It's a shame it seems the authors lost insterest in this project. |
Unfortunately not, but I've found a way to get pretty similar results to highres coloring. I do the coloring on lower resolution and overlay it over the highres bw image using suitable color blending filters. |
@kepppis Yeah, that's a classic trick. =) Unfortunately the "color mask" is still too blurry to my taste, color details are lost {eyes, lips, etc.} and colors are bleeding. I wondered if it would be possible to force a higher working resolution and maybe retrain the model but I guess I'll start from a different place. |
Let me know if you find anything usable. |
Well, I tried EbSynth for that and I use my own tool Tinterr (eriador.condak.cz/tools.php), but to get a decent result you have to feed it with a LOT of keyframes... |
My test result (about 1560 frames and 23 key frames) can be seen here: None of this methods is perfect... |
Is it possible to have instructions how to use PNG-Tinterr ? |
Yes, but I have to create a provisional English manual first... I should have created it a long time ago. =] |
@kepppis Well, here is something that might help you: |
I've run couple times into the situation where the coloring is perfect if mindim is less than 450, but with larger values the colorization will fade and the output is with pale colors which are very far from reference images. Tested this with two environments, one with Nvidia Tesla k40 with 11 GB ram and pytorch 1.2.0 (the last version compatible with oldish Tesla card).
Other with Nvidia Quadro P5000 with 16 GB ram and pytorch 1.8.1. The source video and the reference pics are 576p and the only thing I am changing is the mindim value.
The text was updated successfully, but these errors were encountered: