QuickHitRecord
Veteran
Samuel, what app are you using to process the 5D Raw MLV files in your test? I'm finding that this can make a huge difference to the resulting DNG files.
My 2014 camera shootout: a7s, GH4, 5D3+RAW, BMC, pocket...
- Thread starter Samuel H
- Start date
This is great, i'd love to see what the worst / best dynamic range difference looks like. Love my Gh4 but glad I renewed my NeatVideo license.
You know, with the gh4's resolution you could add a lot more noise reducton than the others, making it more even I suppose.
You know, with the gh4's resolution you could add a lot more noise reducton than the others, making it more even I suppose.
Samuel H
Major Contributor
IIRC, it was raw2dng or something like that. The one that was linked in the ML page, that works by dragging the .raw file into the .exe to get the .dng files.
In-camera noise reduction was off. For the NR numbers I applied MB Denoiser II in post, maxed out (300%).
That was my point for using NR on the GH4 only: it loses some quality (not sharpness, but the overall look), but even with a ton of NR that 4K image is quite nice.
Well, I found that the GH4 can match the 5D3+H.264 in the highlights, but it doesn't go that far into the shadows (well, it goes there, but with a ton of noise, thus the need for NR).
In any case, even when you get those shadows back in the GH4, and even with CineLike D and contrast=-5 the image is really contrasty compared with what you can get with some of the other cameras. If you're going to grade those shadows, you probably want 10-bit recording, otherwise the gradation is going to be pretty poor.
Erik Wittbusch
Well-known member
THX for the test!
BMCC and BMPCC use exactly the same sensor but different sizes.
If you process both the same way, the DR, color, tonality should the same.
Only FOV is different. (BMCC produces ProRes by downsizing the original 2,5k
footage to HD and is therefore sharper).
This is the same with all cameras. The post processing and the picture profiles
are very important for the final result! Personally I don't like the ML RAW workflow
for bigger projects and I don't like the weird color, the A7s produces by using
SLog2 gamma without the appropriate color space.
My findings differ from Samuel's but are clearly not closer to reality but give
results depending on my personal post workflow.
BMCC and BMPCC use exactly the same sensor but different sizes.
If you process both the same way, the DR, color, tonality should the same.
Only FOV is different. (BMCC produces ProRes by downsizing the original 2,5k
footage to HD and is therefore sharper).
This is the same with all cameras. The post processing and the picture profiles
are very important for the final result! Personally I don't like the ML RAW workflow
for bigger projects and I don't like the weird color, the A7s produces by using
SLog2 gamma without the appropriate color space.
My findings differ from Samuel's but are clearly not closer to reality but give
results depending on my personal post workflow.
Samuel H
Major Contributor
I have no idea how the BMC result came to be, but it is "original BMC" shooting 2.5K RAW, and I used exactly the same settings in both cameras. I also expected them to have the same result, since the sensor is so similar. Still, the pocket came out ahead both in the highlights and in the shadows (and yes, the difference in shadow noise was obvious).
Maybe it was the lens? Though it would still be surprising, since it's a big difference for such similar lenses (both Leitz vintage, Elmarit-R 35mm f/2.8 set at f/8 on the original BMC, Summilux-R 50mm f/1.4 over bmpcc speedbooster on the pocket, set at f/13).
My best guess would be that the sensor fab is doing better now than it did a couple of years ago, since the BMC was one of the very first models shipped (first one to reach Spain, I was told), and the pocket is much more recent (around 6 months old). Or maybe it was just good/bad luck for any of them, who knows.
RE: color issues on a7s with slog2: it's not slog2, it's sgamut; use Flaat instead.
Maybe it was the lens? Though it would still be surprising, since it's a big difference for such similar lenses (both Leitz vintage, Elmarit-R 35mm f/2.8 set at f/8 on the original BMC, Summilux-R 50mm f/1.4 over bmpcc speedbooster on the pocket, set at f/13).
My best guess would be that the sensor fab is doing better now than it did a couple of years ago, since the BMC was one of the very first models shipped (first one to reach Spain, I was told), and the pocket is much more recent (around 6 months old). Or maybe it was just good/bad luck for any of them, who knows.
RE: color issues on a7s with slog2: it's not slog2, it's sgamut; use Flaat instead.
Erik Wittbusch
Well-known member
THX for the additional info.
I will keep away from the A7s as an A-cam and bought the FS7 instead.
It uses SLog3 with SGamut3.cine and gives nicer results. But at the same time my pockets
are more than empty now...
SGamut is made for different sensors (all of them use at least 10bit codecs) and it seems that
it's huge color space is not good for 8bit anyway as is SLog2.
When I tested the A7s, I found the cinegammas to work best because the tonality and color came
out nicer than SLog2.
as always: YMVV
QuickHitRecord
Veteran
When you were shooting the 5D3 footage, did you use the ETTR hint for exposure?
Erik Wittbusch
Well-known member
FYI:
SGamut is always the same - call it SGamut2 or SGamut3.
SGamut3.cine is the new narrower color space...
SGamut is always the same - call it SGamut2 or SGamut3.
SGamut3.cine is the new narrower color space...
Samuel H
Major Contributor
I chose exposure for all the cameras looking at my light meter. I lit the scene for f/11 ISO 1600 using the grey card you can see over the spydercheckr, and I set the 5D3 to f/2.8 ISO 100, the BMC to f/8 ISO 800, etc. (the only exception is the a7s, which I always overexpose by around one stop, so that one went to f/11 ISO 3200).
I used f/ instead of T/ but that shouldn't be a big problem since the lenses are so similar (and in many cases exactly the same one, e.g. 5D3 and a7s used the same 90mm lens).
Samuel H
Major Contributor
Sorry, the sgamut3 cameras are out of my reach so I don't know the correct terminology. All I know is that some more expensive cameras have a linear color space called sgamut3 or something like that (probably the sgamut3.cine that you mention), whereas my a7s has a version of sgamut that introduces significant color non-linearities (skin tone changes hue quite visibly depending on luminance levels).
Erik Wittbusch
Well-known member
The problem of the SGamutX color space is its size.
It's huge and has a big range esp. in the greens.
This is okay for 12bit RAW or 10bit 4:2:2 but on 8bit codecs
it's not really possible to bring it back to "normal" (Rec709)
without big problems.
All log footage must be delogged with the appropriate LUT
before grading. That's what most people forgot when they wanna
use log because it's cool and gives lot of DR.
A log/LUT workflow is required when shooting this way and this is
not really made for 8bit limited bandwidth...
It's huge and has a big range esp. in the greens.
This is okay for 12bit RAW or 10bit 4:2:2 but on 8bit codecs
it's not really possible to bring it back to "normal" (Rec709)
without big problems.
All log footage must be delogged with the appropriate LUT
before grading. That's what most people forgot when they wanna
use log because it's cool and gives lot of DR.
A log/LUT workflow is required when shooting this way and this is
not really made for 8bit limited bandwidth...
Samuel H
Major Contributor
I adjusted my measurement of the usable DR for the a7s, taking into account that colored highlights look awful about a third of stop before cliping (I measured that it's just a third of a stop, and now my number is 11.7 stops instead of 12 stops).
I also posted full shots with every camera, so you can see how that difference looks like. Hint: one stop can be a lot...
here
I also posted full shots with every camera, so you can see how that difference looks like. Hint: one stop can be a lot...
here
niki
Active member
great test…can you include a strip of kodak 5203 next time..it's unbelievable
starcentral
Veteran
In order to eliminate any large or small lens characteristics when truly comparing cameras is there any reason you did not wish to 1.) use the same lens for all tests, 2.) keep the lens aperture constant and the same from camera to camera through use of ND filters and/or possibly adjustment of lighting if required?
Samuel H
Major Contributor
I went for different lenses because I wanted to have a similar FoV with all cameras. And the optimal ISO in each case varies a lot too, which is what made me use a different aperture in each case. I think this was the right thing to do for us because the reason we ran this test is that we have to choose which camera we use for an upcoming project. Using a different lens and ISO on each camera made sense because it's what we would have to do when we shoot.
Where you're right is that I should have used ND filters to bring aperture a bit closer to each other (can't match them, my set is sparse, I just have filters for 3, 6 and 9 stops; but with that I could have kept all cameras between f/4 and f/8, getting rid of the odd f/2.8, f/11 and f/13 results) (in fact I wouldn't want to match aperture, but DoF, which means comparing f/8 on the a7s with f/4.5 on the pocket+SB; I could have done this -approximately- with ND). Using ND thing would have been a close substitute to what we would do in actual shooting (which would be: bring in smaller lights).
In any case what I was most interested in was DR, and that doesn't depend so much on what lens and aperture you use.
Where you're right is that I should have used ND filters to bring aperture a bit closer to each other (can't match them, my set is sparse, I just have filters for 3, 6 and 9 stops; but with that I could have kept all cameras between f/4 and f/8, getting rid of the odd f/2.8, f/11 and f/13 results) (in fact I wouldn't want to match aperture, but DoF, which means comparing f/8 on the a7s with f/4.5 on the pocket+SB; I could have done this -approximately- with ND). Using ND thing would have been a close substitute to what we would do in actual shooting (which would be: bring in smaller lights).
In any case what I was most interested in was DR, and that doesn't depend so much on what lens and aperture you use.
Last edited: