"Dark Matter" - Exploring MLVapp feature RBF denoiser

Started by ML700D, February 09, 2022, 05:34:28 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

ML700D



trying to shoot in low light and using RBF denoiser to pull out a bit detail in the shadow.
it's hard but worth trying.
EOS 700D

Skinny

Have you tried uploading the same video to youtube but without denoiser? It will be interesting to compare. I found that youtube heavy compression algorithm acts like a denoiser to some extent :D

ML700D

EOS 700D

ML700D

Quote from: Skinny on February 09, 2022, 07:09:45 AM
Have you tried uploading the same video to youtube but without denoiser? It will be interesting to compare. I found that youtube heavy compression algorithm acts like a denoiser to some extent :D
interesting.. 🤔😄


EOS 700D

Skinny

So, what do you think?  :) It will also look better with stabilization, because youtube is especially lazy when some parts of a frame don't move, youtube considers them static and just replaces with permanent picture. and it also increases quality of small details.
This is not scientific or something, just some of my observations. At least now I don't really care about small amounts of noise, in 1080 or less nobody will see it. Or, at least, there is no need for extreme denoising..

masc

For most jobs I think RBF denoiser is good enough. Here my example for 5D3 footage using ISO1000..4000 + Darkframe subtraction + RBF denoiser + YT denoiser ( sry, compression ;) ).
https://youtu.be/69zvr4VFkSw
5D3.113 | EOSM.202

ML700D

Quote from: Skinny on February 09, 2022, 03:31:31 PM
So, what do you think?  :) It will also look better with stabilization, because youtube is especially lazy when some parts of a frame don't move, youtube considers them static and just replaces with permanent picture. and it also increases quality of small details.
This is not scientific or something, just some of my observations. At least now I don't really care about small amounts of noise, in 1080 or less nobody will see it. Or, at least, there is no need for extreme denoising..

yes.. you are right.
EOS 700D

ML700D

Quote from: masc on February 09, 2022, 06:25:20 PM
For most jobs I think RBF denoiser is good enough. Here my example for 5D3 footage using ISO1000..4000 + Darkframe subtraction + RBF denoiser + YT denoiser ( sry, compression ;) ).
https://youtu.be/69zvr4VFkSw

wow.. impressive masc!

btw what is darkframe subtraction? thanks.
EOS 700D

masc

5D3.113 | EOSM.202

Skinny

@masc those clips look very clean. It is hard to believe they were shoot at 1000-4000 iso. But 5DIII is much better in low light than other ML cameras..
When you first posted this video I also noticed jumping lens flares, so you must be using software stabilization? I thought a gimbal will provide very smooth footage, but how steady it is in reality?

ML700D

Quote from: masc on February 10, 2022, 12:37:32 PM
MLVApp offers to reduce the static noise from your clips using dark frames. https://github.com/ilia3101/MLV-App/wiki#darkframe-subtraction

thanks masc.

already subscribed your channel. 🍻
EOS 700D

ML700D

in that web said..

QuoteRecord with identical settings and identical equipment a clip with lens cap mounted. Some seconds are enough.

masc, do I have to make dark frames every time I change the setting?

e.g. record with iso 800 f/2.8 then I use the same setting with lens cap mounted to make dark frames, then I recorded other with iso 400 f/1.8.
EOS 700D

masc

Quote from: ML700D on February 10, 2022, 02:51:59 PM
do I have to make dark frames every time I change the setting?
e.g. record with iso 800 f/2.8 then I use the same setting with lens cap mounted to make dark frames, then I recorded other with iso 400 f/1.8.
Yes. That is correct. It is the only way it can work.

Quote from: Skinny on February 10, 2022, 02:23:40 PM
@masc those clips look very clean. It is hard to believe they were shoot at 1000-4000 iso. But 5DIII is much better in low light than other ML cameras..
When you first posted this video I also noticed jumping lens flares, so you must be using software stabilization? I thought a gimbal will provide very smooth footage, but how steady it is in reality?
Yes, 5D3 is much better than EOSM and also a very little bit better than 5D2 in terms of noise.
I stabilize all my footage with FCPX. Without additional digital stabilization in post the footage made with my Weebill S is just unusable. ;) You can't move your hands smooth enough to not have shaky footage. At 16mm it can be okay with some luck, but at 50mm it is very noticable. Maybe this becomes better with IS lenses, but I don't have any to try out.
5D3.113 | EOSM.202

ML700D

EOS 700D

ML700D

masc!
I noticed weird thing next to that person in your video. what is it masc? 😂

EOS 700D

Skinny

Quote from: ML700D on February 10, 2022, 02:51:59 PM
then I use the same setting with lens cap mounted to make dark frames, then I recorded other with iso 400 f/1.8.
and also at the same ambient and camera temperatures, if you can. for example, dark frame made with identical settings but with cold camera is absolutely different than when the camera is working for an hour or so. winter/hot summer conditions also should affect temperatures and noise patterns..

I did some experiments with timelapses (1-2h with 1..1.5fps) and if I record the dark frame before the timelapse, it worked good only with first ~1/3 of a video, and if I did dark frame after the timelapse (when sensor is hot) then the beginning of a video looks bad. So I couldn't decide what darkframe should I use.
But then someone suggested that I can export two video variants with both dark frames, and crossfade between them in editing software. Should work, but I have not tried it yet. Of course, for short clips it is not necessary.

Quote from: masc on February 10, 2022, 03:17:17 PM
I stabilize all my footage with FCPX. Without additional digital stabilization in post the footage made with my Weebill S is just unusable. ;) You can't move your hands smooth enough to not have shaky footage. At 16mm it can be okay with some luck, but at 50mm it is very noticable. Maybe this becomes better with IS lenses, but I don't have any to try out.
I see.. bad news, I thought if I buy a gimbal someday, then digital stabilization won't be necessary.  :(

ML700D

Quote from: Skinny on February 10, 2022, 06:13:48 PM
and also at the same ambient and camera temperatures, if you can. for example, dark frame made with identical settings but with cold camera is absolutely different than when the camera is working for an hour or so. winter/hot summer conditions also should affect temperatures and noise patterns..

I did some experiments with timelapses (1-2h with 1..1.5fps) and if I record the dark frame before the timelapse, it worked good only with first ~1/3 of a video, and if I did dark frame after the timelapse (when sensor is hot) then the beginning of a video looks bad. So I couldn't decide what darkframe should I use.
But then someone suggested that I can export two video variants with both dark frames, and crossfade between them in editing software. Should work, but I have not tried it yet. Of course, for short clips it is not necessary.

I see.. so basically record dark frames before/after the shoot, right?
EOS 700D

Skinny

maybe somewhere at the middle, and try what will be looking better


belalkhansocial

Quote from: masc on February 09, 2022, 06:25:20 PM
For most jobs I think RBF denoiser is good enough. Here my example for 5D3 footage using ISO1000..4000 + Darkframe subtraction + RBF denoiser + YT denoiser ( sry, compression ;) ).
https://youtu.be/69zvr4VFkSw

This is simply beautiful!


ML700D

nice.. but what is the process?
for iso 1600 - 3200 in low light will get some noise I think.
EOS 700D

Skinny

from youtube comments:

QuoteSigma 18-50 f 2.8-4.5
24 fps 1600 (wide angle shots) and 3200 Iso
5kFrtp 16/9 mode (3840/2160) 10bit
Export in Arri log C gamma
Graded in Davinci Resolve 17.4

Chek out this video, I'm use this technique for denoise my footage  https://youtu.be/NKC4Me_2pfY



and he also mentions that he used 1/50 shutter speed all the time, but in reality you can easily get away with 1/40...1/30 or even 1/25 with static shots where you don't see a lot of motion. nobody will really see a difference in motion blur, but for low light 1/25 is just +1 stop of light. could be more important than "proper" shutter angle.

ML700D

I see, thanks.
unfortunately DaVinci is heavier than premier on my pc.
I think rbf denoiser is enough for me at the moment.
EOS 700D