I don’t remember what I did with the fine settings anymore, but I think I tried that slider as well. The noise came back before the sharpness was mitigated and I couldn’t find just the right setting.
I’m trying to get rid of the film grain that’s appearing across the screen in the original video, but it seems to be pretty stubborn.
I tried Gaia-CG in VEAI (1.6.0), but it didn’t completely remove the grain noise.
If you set Theia to an extreme setting, the noise can be suppressed, but it is not as clear as Gaia-CG. Both of them have their own merits and demerits.
Looks like your sample is a low-bitrate encode, with artifacts resulting from film grain that was poorly handled. Based on the picture you posted I suppose you encoded the sample quickly instead of cutting straight from the original, which makes it impossible to try anything with it.
Looking at the picture I’d say you would be much better off running a good denoiser like MCTemporalDenoise first. VEAI does not replace a good filter chain and should be your last post-processer.
As you pointed out, the video is already encoded to a low bit rate (roughly 5000Kbps), but it is a lossless cut of the video I’m testing with VEAI, so you should be able to see the equivalent if you download it.
I’m going to learn and try out the pre-processing with the denoiser, but hopefully one day I’ll be able to get satisfactory results from VEAI processing alone.
I wasn’t saying don’t ever use anything but 1/2/4 - I use odd ones all the time, for example 480 to 1080, but rather that when you are showing an issue with VEAI it is best (again, this is my opinion) to use the stock enlargements so you aren’t including after-AI resizing artifacts.
I’m mostly using Artemis HQ for low resolution 320x160 mpeg1 videos and Gaia HQ as the 2nd final-pass for a sharp clean image.
Time to test it with my new RTX 3090
I had problems with Lagarith as imput in older versions of VEAI.
Some frames repeated on longer videos or strange fame jumps occured.
Maybe it was because of the Null-frames coding option?
Since that I avoid that codec. I use now FFV1 or Huffyuv without any problems.
The fact that the sample you uploaded is a MKV and not a M2TS, coupled with its odd dimension (cropping I suppose) tells me that it is not the original. I do not recognize the source but there is no way this is coming straight off a BluRay.
In any case, here’s a frame denoise with a medium-strength MCTemporalDenoise. Try to enhance it and see if you’re getting better results. A smart sharpening filter should do the trick.
Application is Open ? What do you mean - how could I drag and drop anything from a program that is not open, to one that is not open ?
Appeared to work once, I don’t remember when but failed from managers (XnView and FastStone). I have now tested again and i1.70 fails everytime for me. Worse yet the program often does not properly fit the screen but is shifted to the right so I cannot see the menus on the far right. Reloading usually fixed the problem, but sometime it takes 2-3 reloads.
In any case I’m going to load v 1.71 and see if the problem is fixed.
You are correct in your estimation.
I think it was originally a 1.85:1 aspect ratio film, but I cropped it (slightly unsuccessfully) because the Blu-ray had a quirky screen size with black bands on both sides as well as the top and bottom.
It is indeed more natural than the one using Theia’s denoise.
I will look into MCTemporalDenoise and use it. Thank you for your guidance.
No problem. I recommend using DGIndexNV (will only work with a somewhat recent nVidia GPU though) to index the original M2TS and work from here with AviSynth. Feel free to message me if you need help with that.
I didn’t see any after-AI-resizing artifacts, to be frank, but I also didn’t test, if anything like this occurs. I’ll take a deep dive into this at some point in time then.
Just installed 1.7.1. Vast speed increase over 1.6.1 with the same video overall. Surprisingly the Vega 64 is essentially equal to the RTX 2080 Super with Gaia-HQ v5. I’m upscaling a power point presentation from 720p → 4k and getting right about .60 sec/frame. I double checked it was actually switching GPUs and it is. With the experimental multi-GPU option it drops down to .48 sec/frame. I’d assumed the Tensor cores in the RTX series would give it a significant advantage but maybe they don’t. I’m controlling this rig remotely right now but when I get home in a week or so I’ll see how the Radeon VII’s do.
I’m trying the Radeon Pro VII two weeks long right now, but its driver is full of bugs (may release was a paperlaunch). And i did already test the 3090 in Denoise via remote workstation and it was 72% faster than the vega 20 radeon pro VII. But in VEAI as you said, Vega 64 has the same speed as 2080S.
im watching your image on a 150 inch projector screen, and i usually dont like using theta, I usually prefer gaia 90 percent of the time and artemis LQ for some cases. but on this particular conversion you did theta high fidelity actually looks good. on your sideby sides i would say theta high fidelity is number 1 and giga pixel came in 2nd place. do you mind telling me the specific settings you used for theta on this one?
