recent update Reference frames "ColorMnet" video colorization working good . without any error . 1st image original image from video 2nd image refernce based colorization
I just released the new vs-deoldify version v4.6.0
In the new version I added 2 new color models under DDcolor: Real-Time User-Guided Image Colorization with Learned Deep Priors (Zhang, 2017) Colorful Image Colorization (Zhang, 2016)
these 2 models has been added as additional models: siggraph17, eccv16, in DDcolor since have in common the same settings and tweaks.
On first use the following pytorch models will be downloaded:
To add the new models in Hybrid GUI is just enough that you add the new DDcolor models 2 & 3 as described in the description of parameter DDcolor-model contained in the ddcolor_p list:
DDColor-model to use (default = 1):
0 = ddcolor_modelscope,
1 = ddcolor_artistic
2 = colorization_siggraph17
3 = colorization_eccv16
I'll try to look into it.
A bit confused.
a. if this is a modification of DDColor, shouldn't DDColor get adjusted?
b. if this has nothing to do with DDColor, shouldn't the parameters&co get renamed?
Cu Selur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
To simplify the development on my side, and I hope also on your side.
From the point of view of the GUi are just 2 additional models other that "modelscope" and "artististic".
In the HAVC_ddeoldify when are used the model 2 and 3 in ddcolor_p, like in:
will use the new "Colorizer" code and not DDcolor.
The 2 new models have the same instability problems than DDcolor, in this way I can use the filters already developed to stabilize DDcolor, to stabilize also the siggraph17 and eccv16 color models.
and assumed you just added models.
But you also changed the old models.
And changed the defaults, thus invalidating any old settings for this in Hybrid.
Changed it now to: