GH5 really weird noise when doing VFR on GH5s !!!

C

coutard

Well-known member
making another thread about my recent tests with my new GH5s

just got out to make some VFR test this afternoon and looking at the footage I am HORRIFIED with some really weird noise and artefacts in dark levels, when doing VFR only. regular speed footage looks fine

examples : (download them to see accurately)
https://www.dropbox.com/s/kx008lga4l79507/test3-VFR-GH5s.mov?dl=0
https://www.dropbox.com/s/q26t5u45k3a6et2/test2-VFR-GH5s.mov?dl=0
https://www.dropbox.com/s/m5g0jtkxjcj6dq2/test-VFR-GH5s.mov?dl=0

from what I've seen online it doesn't look right !

shot in HD 4:2:0 8bits V-log at 640 ISO

any thoughts ?
 
The major problem with it is simply low bit rate h.264 encoding 14.8 mbps, no CABAC entropy, 8 bit 4:2:0. If you doubled your encoding bit rate, it would go a long way to fixing that. You are also encoding the noise floor of raised exposure, or video levels. In other words, it looks like it is recorded or transcoded (can't be sure which) at 16-255.

Bringing it into a Resolve YRGB color managed timeline and setting the input space to Panasonic V-gamut/VLog, the recording levels (or your encoding levels) appear to be 16-255. By using the Resolve log control "offset" to bring down the levels it is reducing the noise floor and cleaning up the dark areas properly. Then what you are left with is two remaining issues, low bit rate; spending your low rate bit budget encoding motion on the water and non-corrected chromatic lens aberrations, and the noise floor which you are also wasting bits encoding by having the noise floor raised above 16 into the visible range of code values.

It all is easily correctable and to me otherwise looked like it could be fine once you get your workflow straightened out.
 
another exemple : https://www.dropbox.com/s/inbc0zxz9pqv39v/test4.mov?dl=0

I really don't see it as a bitrate issue. my C100 at 24mbps didn't have this problem, even in log.
plus, I don't see why you're talking of 14.8mbps since the camera is recording at 100mbps for this mode ?

And, even if I could "hide" it with lowering the blacks, this is not a way of doing that I like. footage should not look like this. Very often I deliver my shots to the production and I don't have control over post-production, and I really couldn't deliver those kind of shots without looking like a complete amateur

I am worried that my version could be faulty ...
 
ah you were talking about the compression I made for uploading. Yes it is h264 limited to 15mbps but its not the point, I did this for uploading reasons, the problem is exactly the same with original files
 
I'm seeing exactly what I would expect from 8-bit 4:2:0 V-LOG L. H.264 is designed to compress the hell out of shadows - with a milky image like this, it's fairly noticeable.

- You'll have better luck with a rec.709 profile. Cine-D would be a good choice if delivering with other V-LOG L footage. Natural is also a strong choice is delivering standalone.
 
I really don't see it as a bitrate issue. my C100 at 24mbps didn't have this problem, even in log.
Click to expand...

Then upload an original. It's not about hiding blacks, I agree. It is about having the correct exposure, and making sure your upload did not change the levels from the original. You didn't mention your gain/iso setting, your exposure, just that your C100 doesn't do that, leads me to a conclusion I can't be certain about anything here right or wrong. No detail specified what is wrong with it, just posted clips and mention to the C100. If you were going down that path, I would chip in that my 5DmkIII managed shadow noise by just crushing everything before it recorded the track. There was no noise and no detail in the shadows. My F55 would show shadow noise in S-Log3, a consequence of capturing more detail. More DR means is going to have more noise at the bottom, your particular tolerance for it and what you do with it in post is what makes defining DR so ambiguous. You said, hey look, I did look. On the surface just seems typical of lots of log footage, not all of it good, nor bad.

Canon C-Log is not the same as V-Log/L, I don't know if you were exposing for middle gray, ETTR, but that's just how log footage works, tones are compressed toward the middle making room at the ends, so they have to lutted, blacks and whites and skin tones and middle gray all get moved around and it's not "hiding" anything.

I'm not saying you are wrong either, but the vagueness at this point has not persuaded me that it's not workflow related.
 
V-log in 8bit is sopposed to be nasty. What does it look like in CineD or one of the other profiles.
 
here is two orignal files : https://mab.to/aeKDEnqs5

you'll see 2 files : 25p and slow-motion 100fps. both at 2000 ISO in vlog.

There were no vagueness in my post. I mentioned 640 ISO ; by looking at the clips you notice they're not underexposed (even if they were, problem is still something else) and I would have been silly of me to upload low bitrate h264 that showed different problem to the orignal. Thought it was implicit.

I know what log is. I know it raises the blacks. I know vlog is different to c-log. I know how to expose my shots because its been my job for 10 years and I've been delivering footage for TV and cinema, so, no workflow issue.

I'm just saying that regarding my experience, log footage shouldn't look like that. I did a doc recently with a F55 like yours, in slog3, it was not at all looking like that, even in low light

I know its 4:2:0 8bits but again, I think its not the point. log is supposed to raised blacks, right, but not add noise to it. when in low light and underexposed situations, yes, I agree, log is not a good option. But here, middle of a sunny afternoon, no, I don't buy it

And I don't intend to change my PP to cine4 or anything else, because i'm supposed to deliver V-log.
and still, I don't like avoiding a problem by looking the other way, I want to understand why it does that !
 
VFR on the GH5 is really weak. Your spreading 100mbps across too many frames and as a result the footage is just kind of shitty. Again try profiles meant to be used in 8bit and see if there is a difference. I think you are just expecting too much out of this camera.
 
^^^ What he said.

It's real, probably inherent, cleaned up with minor NR in Resolve, but if you're handing off footage to a client, I don't know what the answer is, tell them about it.

And on the point, I observed what it looks to me like fine grain sensor noise, snow and chroma like what you'd see with higher gain/iso.
 
okay...so to you my body wouldn't be faulty ?

I don't know if I was really expecting too much, it's just that....for exemple before I bought it I looked at this https://vimeo.com/211425610
and I really don't see anything like what I'm getting with mine. Even with NR in resolve and grading, it looks way cleaner....
 
coutard said:
okay...so to you my body wouldn't be faulty ?
Click to expand...

Not faulty.

I saw what was happening, the lens hood in your scene and other dark elements (filter ring) are underexposed because you protected the highlights coming in through the windows from clipping. Or you just couldn't get enough exposure with your combination. The darker elements are visible above the sensor's minimum sensitivity level but below the noise floor, in the region of visible sensor noise.

I copied your GH5s settings, V-Log/L, PAL 50hz, 25 fps playback, 100 fps VFR, and saw the potential for the same noise but the low key in my shot was above the noise floor, staying clean. When I panned into the light, it clipped badly because it was getting more exposure, iris opening was f/2.2 and the scene was adequately lit.

Your VFR clip still graded satisfactory in my opinion, but if you were to lift the level in post to brighten the darker elements to your satisfaction the noise will come up with it. There is a limit on what you can push in post and that limit is your tolerance for the noise. It needs more light.
 
Tom Roper said:
Not faulty.

I saw what was happening, the lens hood in your scene and other dark elements (filter ring) are underexposed because you protected the highlights coming in through the windows from clipping. Or you just couldn't get enough exposure with your combination. The darker elements are visible above the sensor's minimum sensitivity level but below the noise floor, in the region of visible sensor noise.

I copied your GH5s settings, V-Log/L, PAL 50hz, 25 fps playback, 100 fps VFR, and saw the potential for the same noise but the low key in my shot was above the noise floor, staying clean. When I panned into the light, it clipped badly because it was getting more exposure, iris opening was f/2.2 and the scene was adequately lit.

Your VFR clip still graded satisfactory in my opinion, but if you were to lift the level in post to brighten the darker elements to your satisfaction the noise will come up with it. There is a limit on what you can push in post and that limit is your tolerance for the noise. It needs more light.
Click to expand...

I understand what you say but thing is, at exactly same exposure, i.e same levels for lens hood and filter, noise is different between VFR and non VFR. That is my main question.

regarding exposure, of course, the hood and filter were "underexposed". they were not, actually, they're just black ! black is not supposed to fall in the middle of the waveform. and I don't see any reason why I would raise them in post. I would lower them near IRE0, like...blacks.

did some test in CINE-D and some nasty noise still appears in deep blacks. as CINE-D is a bit "graded", it shows less, but in true blacks I can see it and it looks ugly
 
I am not saying you should be satisfied with what you got. In fact I think it is good that you are not. What I am saying is if your camera is faulty, than so is mine and probably everybody else's. The best explanation for the noise I've heard is from sgttom who said it is spreading too many frames across 100 mbps, and that's the difference between the high frame rate and the realtime speed version, with equal exposure. That said, the waveform monitor makes it appear there is 1/2 - 2/3 of a stop more exposure on the realtime clip, even though you properly reduced it by 2 stops, just saying.
 
You must log in or register to reply here.
Back
Top