Posts: 11.185
Threads: 58
Joined: May 2017
With the H100, you should be able to use the torch-addon an ml tools, so scunet&basicvsr++ are an option.
You should make sure to use lossless formats as intermediate otherwise you end up compressing compression artifacts and creating new compression artifacts,..
Quote:(but I just tried Hybrid, and found it better, but the Bob is not working to double frame rates, and I have to increase to 60, to get 30fps. it's not automatically set correctly.)
Then you are doing something wrong. 
Hybrid can bob deinterlaced fine (assuming it knows it should bob and that the source is interlaced).
Hybrid also can interpolate to different frame rates.
Quote:If it didn't zoom in, it would cause the border to be exposed at 100%, because of the shaking and Rock Steady smooth out, so all frames are fitted to size, position, angle, etc., to make it look nice, without reflections or black borders. Simple cropping doesn't work.
Normally, stabilization should add black borders all around, then once crops those borders and if needed/wanted resizes the result. The last resize should be skipped.  So the idea is not to do a simple crop, but deshake and then crop. So if for stabilization you loose XY pixels at each side your resolution is lowered by that.
Quote: We really need to know what we are doing, and would be better if the GPU was used. Max Conversion rate is about 5-6fps.
Hybrid too has tons of encoders and filters which support using the gpu.
Quote:I also use HandBreak for deinterlacing SD,
Personally, I wouldn't use handbrake at all in this process.
Quote:Hybrid is best in serious problems, and the steepest learning curve by far.
Yes, Hybrid assumes that the user knows what he is doing.
Cu Selur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
Posts: 9
Threads: 1
Joined: Aug 2024
(28.03.2025, 22:45)Selur Wrote: Quote:(but I just tried Hybrid, and found it better, but the Bob is not working to double frame rates, and I have to increase to 60, to get 30fps. it's not automatically set correctly.)
Then you are doing something wrong. 
Hybrid can bob deinterlaced fine (assuming it knows it should bob and that the source is interlaced).
Hybrid also can interpolate to different frame rates. I just tried it out for the first time, because converting an old MOV file to MP4, Handbreak did a mountain/forest set, and started meshing the trees badly. I tried many combinations, and finally had it come out perfect. The frame rate went to 15, or half, but I precompensated to 60, to fixed it, and came out better than HB.
Posts: 9
Threads: 1
Joined: Aug 2024
-- removed unnecessary full quote; by Selur ---
I'm taking about the Moire artifact in the blue sky. Looking at the pixel level, it has a comb pattern on the sides, about a quarter of the frame. when the player window is scaled, it's a very noticeable, blocky Moire pattern, not quite Fraunhoffer lines. When the defishing is applied, the pixels are distorted within Mercalli to make the frame look normal. Something Mercalli is doing makes this pattern.
After I applied my filter pipeline, is smooths it out pretty good, most of the time, along with the degrain that I need.
I scaled using Hybrid's viewer to 2x, to test before and after, with each filter, until getting the cleanest result.
I just tried the 10 bit, and the filter is very clean, and seems to have worked in one pass, but do I need all of those 4 filters now? I hope to speed it up.
I can't find what you mean, LumaMask, which filter.
Posts: 11.185
Threads: 58
Joined: May 2017
29.03.2025, 08:13
(This post was last modified: 29.03.2025, 09:26 by Selur.)
Masking controls, and a few others, are hidden by default.
You can show them under 'Filtering->Vapoursynth->Misc->UI'. Enabling options there will extend the controls of most of the filters.
To me these artifacts look like a noise source got compressed to strong which caused compression artifacts which then got again distorted (probably due to the defishing and zooming) and then recompressed creating compression artifacts while trying to keep compression artifacts.
This is why I recommend to first denoise&co and use lossless compression (even if the file size is huge) and then use Mercalli and output again to a lossless format before the final filtering and compression.
Since your source isn't really 4k (doesn't really seem to be 1080p even) at the stage you showed me, you might want to consider lowering the resolution and move all other filtering behind the resizing. ('Filtering->Vapoursynth->Misc->Filter Order/Queue' lets you control the order of the filters.)
Cu Selur
Ps.: using something like FFT3DFilter with Sigma 20 and Neo enabled also is rather fast and shows the distorted grid pattern. (which is why the filtering should be done before distorting the image not after)
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
Posts: 11.185
Threads: 58
Joined: May 2017
A fast approach would be using: - 'Bilateral (GLSL)' with 'Intensity' set to '0.3'
- 'FFT3DFilter' with 'Sigma' set to '20' and 'Neo' enabled
- adjusting the filter order so that 'Bilateral (GLSL)' is before 'FFT3DFilter'
![[Image: grafik.png]](https://i.ibb.co/pr6sp9Vh/grafik.png)
(it would still be better to avoid the creation of those artifacts)
Cu Selur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
Posts: 9
Threads: 1
Joined: Aug 2024
I'm still running Hybrid 2024.06.16.1, I'll upgrade it soon.
Using 10 bit 264, I can take off MLDegrain in the pipeline, and it will run 60% faster. It's near perfect.
I have to test your approach yet.
As far as reducing to 2K, my video editing strategy is to crop my 4Ks to size and position of interest, then output to 2K in Vegas. It's a typical method, because you get 4x the area stored for each clip, and gives the luxury of panning and sizing wherever you want. I don't care if the "4K" res is used, just the 2K crop, and nobody will know the difference. Vegas upreses the images to whatever is needed automatically.
The 4K res is likely limited with the 17mm lens being much less than 8MP, from the factory. I have many hundreds of clips with this lens.
I upgraded to a Macro/Telephoto: SunVision M12 3MP, f=2.8-12mm IR F1.4, Lens Angle: 82-35°(23°)@4K, Zoom: 1.08-2.5x(3.8x), Dia: 28mm, all my new stock is at least 3MP, and nearly undistorted within 1%.
I still have similar grain patterns sometimes with it, but I'm not sure from what, either the camera, and/or Mercalli.
I'll check the raw clips and see.
[*]causes a failure>>> 'Bilateral (GLSL)' with 'Intensity' set to '0.3', Mode=Median, Spacial = 1.0
[*]'FFT3DFilter' with 'Sigma' set to '20' and 'Neo' enabled
[*]adjusting the filter order so that 'Bilateral (GLSL)' is before 'FFT3DFilter'
Vapoursynth preview error: [31mValidation failed: !params->renderable || fmt_caps & PL_FMT_CAP_RENDERABLE (../../../../../src_packages/libplacebo/src/gpu.c:234)[0m [31m for texture: ../../../../../src_packages/vs-placebo/src/shader.c:122[0m
Vapoursynth preview error: [31mValidation failed: !params->renderable || fmt_caps & PL_FMT_CAP_RENDERABLE (../../../../../src_packages/libplacebo/src/gpu.c:234)[0m [31m for texture: ../../../../../src_packages/vs-placebo/src/shader.c:122[0m
Vapoursynth preview error: [31mValidation failed: !params->renderable || fmt_caps & PL_FMT_CAP_RENDERABLE (../../../../../src_packages/libplacebo/src/gpu.c:234)[0m [31m for texture: ../../../../../src_packages/vs-placebo/src/shader.c:122[0m
Vapoursynth preview error: [31mValidation failed: !params->renderable || fmt_caps & PL_FMT_CAP_RENDERABLE (../../../../../src_packages/libplacebo/src/gpu.c:234)[0m [31m for texture: ../../../../../src_packages/vs-placebo/src/shader.c:122[0m
Vapoursynth preview error: [31mValidation failed: !params->renderable || fmt_caps & PL_FMT_CAP_RENDERABLE (../../../../../src_packages/libplacebo/src/gpu.c:234)[0m [31m for texture: ../../../../../src_packages/vs-placebo/src/shader.c:122[0m
Vapoursynth preview error: [31mValidation failed: !params->renderable || fmt_caps & PL_FMT_CAP_RENDERABLE (../../../../../src_packages/libplacebo/src/gpu.c:234)[0m
Vapoursynth preview error: [31m for texture: ../../../../../src_packages/vs-placebo/src/shader.c:122[0m
I used a different clip than I sent.
Posts: 11.185
Threads: 58
Joined: May 2017
Quote: Vapoursynth preview error: [31mValidation failed: !params->renderable || fmt_caps & PL_FMT_CAP_RENDERABLE (../../../../../src_packages/libplacebo/src/gpu.c:234)[0m [31m for texture: ../../../../../src_packages/vs-placebo/src/shader.c:122[0m
Seems like for some reason GLSL filtering doesn't work for you.
Quote:As far as reducing to 2K, my video editing strategy is to crop my 4Ks to size and position of interest
So you end up with 2K content that is like some badly upscaled SD.
Assuming your raw footage from the camera has more details, you might want to look into your workflow, this lack of detail 4k is not normal.
Quote: Using 10 bit 264,...
10bit helps with banding and offers better compression/quality ratios.
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
|