DeepEnhancer

[djilayeden]

Dear Selur and Dan,
I was wondering if it might be possible to consider integrating https://github.com/jiangqin567/DeepEnhancer into HAVC.
This tool provides high-quality video enhancement with a focus on preserving fine details and temporal consistency, which could be a great complement to the current restoration and colorization workflow. It’s particularly effective for improving low-quality or compressed footage—something that often comes hand-in-hand with old black and white films. Having it in HAVC would give users a seamless way to not only colorize but also enhance the overall visual quality of their source materials.
It could be a powerful addition to the toolkit you're building—especially with the momentum HAVC has right now.
Thanks again for your incredible work and your openness to new ideas!
Best regards,

[Selur]

Does it actually work? I'm a bit sceptical of 1-commit projects where nothing hast happened since the inital commit,...

[djilayeden]

Dear Selur,
I wanted to share that I successfully tested DeepEnhancer under Windows Subsystem for Linux (WSL), and it worked perfectly! I installed all the requirements manually, and everything ran smoothly without any issues. The results are honestly stunning—visually impressive and very promising for old footage enhancement.
If you run into any trouble downloading the pre-trained model, feel free to let me know—I have it saved on my Drive and can share it directly.
DeepEnhancer could be a great addition to HAVC and a powerful complement to the existing tools.
Best regards,

[zspeciman]

Hi djilayeden, DeepEnhancer does look visually stunning on their github demo, good find.  It would be nice to see a second Demo.  Please share a short video clip if you can of your own demo; before and after.

[Selur]

Good to know that it at least works. I agree a before&after not from their page would be interessting.
To integrate it into Hybrid, a Vapoursynth wrapper would be needed. (my guess is, that this wouldn't be too hard for Dan64)
But from what I read, this probably should be a separate filter and not a part of HAVC.

[djilayeden]

I'll do it tonight! thank you

[safshe]

@djilayeden
Could you please share the pretrained model on Google Drive or another cloud storage service? I am unable to download it from Pan.baidu.

[djilayeden]

Dear Selur,
Here is my test with DeepEnhancer:
👉 https://drive.google.com/file/d/1tB5DuhxqRiMko_1OpBeLG3oSUC9YhElh/
The visual improvement is quite impressive—details are sharper, noise and spots are reduced, and it truly gives new life to old footage. This should give a more realistic idea of how it performs in real-world usage.
Thanks again for everything!
Best regards,

[Dan64]

Does it actually work? I'm a bit sceptical of 1-commit projects where nothing hast happened since the inital commit,...

Please Selur, move this post and all the related posts in a new thread, maybe titled: " DeepEnhancer"
I'm not interested in this model because is not related to coloring images and I will not add it in HAVC.

Moreover I'm not impressed in the restoring capability, in my opinion in Hybrid there are better tools (*).

I was not able to download the trained model because I don't want to install Baidu disk manager (**).


Dan

(*) It is based on Basic Super Restoration
(**) The simpler way is to add the trained model as released asset in github, but the author preferred Baidu...

[Selur]

As requested, I split off the DeepEnhancer related posts.
@Dan64: I agree, doesn't seem to impressive. Spotless and some target filtering would probably do better.

Cu Selur