Dynamic range tests

[Andrew Reid]

Let's shoot some quick a dirty dynamic range tests

Here's mine, on the balcony

First is Sigma Fp-L in 8bit Cinema DNG RAW

Second is OM System OM-1 in 10bit LOG 4K:

A clear win for the Sigma Fp-L

You will need a bigger effort to get the best out of the OM-LOG

The Fp-L RAW seems to have that creamy RED EPIC / ALEXA look to me straight out of the can

And it doesn't sacrifice a boat load of colour to achieve such a wide DR.

This was in full frame mode to SD card.

I am sure the external 10bit and 12bit is even better in the highlights.

[hyalinejim]

That's pretty ouchy for the Olympus colour. Is it still in its native log gamut or has it been converted to Rec709?

[Andrew Reid]

Native LOG gamut, I can get it looking a lot better but wanted to a quick and dirty look at the files with minimum of corrections. The DNG RAW just comes out lovely with no effort.

Also I feel it has that "digs deep in the shadows" look that cinema cameras have whereas F-LOG and OM-LOG look like animated GIFs in the shadows, and too much in way of crushed blacks.

[MrSMW]

I’m on my phone but clear win for the Sigma…as I’d expect and the colour looks pretty decent too.

I’ll take a look later on my PC and big screen…

[MrSMW]

On my PC, night and day.

But I'd expect that...

It's actually very useful because it pushes me further away from any consideration to ever going back to 4/3rds as much as I have a very decent business case for an OM-1 or GH6 based kit.

And at the same time...you twat, makes me more interested in the FP-L again 😜

As I mentioned the other day, Ninja V hinged at the rear plus 24-105mm f4 for OIS and that could be a pretty fabulous, still pretty compact cine beast...

Not DR related, but how's the electronic stab on the FP-L and how is it in combo with the OIS of the 24-105mm?

I know they do not directly play together like say Panny's IBIS and OIS, but do they work together at all? I'm not bothered by a bit of cropping as I tend to shoot longer anyway.

One other option that makes the proposition even more tasty, but slightly different and that is with the Fotodiox EF-L mount adapter with VND between the lens and body, EF mount has a lot more OIS lens options...

Damn damn damn you Reid, you have my brain pondering again... 

[deezid]

That IMX410 sensor with 12 bit readout (also used in that configuration in the FX9, Panasonic S Series and Zcam F6) is pretty much unrivalled when it comes to color rendering and latitude in its price region.

 Only downside is its slow readout though.

[Eric Calabros]

15 minutes ago, deezid said:

when it comes to color rendering

Don't want to be "akshually" guy but.. If its RAW, it has no color rendering. Color is what RAW developer software provides. Image sensor doesn't output a "look".

[Andrew Reid]

That's the problem with LOG. It always has a look baked in and it takes serious effort to get the best out of it and tailoring to a particularly scene. Whereas with RAW it is so much easier to get it looking cinematic. Magic Lantern RAW on the 5D Mark III in 3.5K still ranks above 95% of the camera market in terms of just image quality.

Will do a test later in low light and see if the Fp-L holds up at high ISOs because it is a 9.5K sensor. The most data at 24fps it can do is 7K crop at 30ms rolling shutter, but the pixel binned 4K from full frame 9.5K has held up really well so far, it doesn't really have any moire or aliasing and dynamic range is fully intact.

[Avenger 2.0]

2 hours ago, Andrew Reid said:

That's the problem with LOG. It always has a look baked in and it takes serious effort to get the best out of it and tailoring to a particularly scene. Whereas with RAW it is so much easier to get it looking cinematic. Magic Lantern RAW on the 5D Mark III in 3.5K still ranks above 95% of the camera market in terms of just image quality.

And those 5D Mark III become dirt cheap now. Just bought myself a 5D3 with 300k clicks and some battery/memory cards for just €200 😎

[mercer]

7 hours ago, Andrew Reid said:

Let's shoot some quick a dirty dynamic range tests

Here's mine, on the balcony

First is Sigma Fp-L in 8bit Cinema DNG RAW

Second is OM System OM-1 in 10bit LOG 4K:

A clear win for the Sigma Fp-L

You will need a bigger effort to get the best out of the OM-LOG

The Fp-L RAW seems to have that creamy RED EPIC / ALEXA look to me straight out of the can

And it doesn't sacrifice a boat load of colour to achieve such a wide DR.

This was in full frame mode to SD card.

I am sure the external 10bit and 12bit is even better in the highlights.

This looks really good. I always suspected the internal 8bit raw was usable. I'd imagine it would do well with a processed noir B&W image.

[mercer]

2 hours ago, Andrew Reid said:

Magic Lantern RAW on the 5D Mark III in 3.5K still ranks above 95% of the camera market in terms of just image quality.

Agreed. Even the stable 1080p still pulls its weight over most $3000, and under, cameras... especially with the right lens combo.

Throw an old non-ai Nikkor, like the 28mm f/2, on it and it will give an almost instant 70s/early 80s gritty film look like something from Mean Streets.

[mercer]

3 hours ago, Eric Calabros said:

Don't want to be "akshually" guy but.. If its RAW, it has no color rendering. Color is what RAW developer software provides. Image sensor doesn't output a "look".

No, but the camera's filter array will. For instance, Raw video from a Canon will have its own look compared to other cameras, no matter how it's processed.

Well... sure you can make it look however you want, but a straight conversion would yield a different color look than a straight conversion from another camera's raw video.

[Andrew Reid]

Yeah sensors have their own look in the RAW data.

The filter stack has an effect.

The CFA and pixel design makes a difference to colour as do better know traits like signal to noise ratio and dynamic range.

I shot Digital Bolex and 5D Mark III at same time once, same uncompressed RAW output in 1080p at same frame rate.

They looked totally different in terms of colour!

And you do have to process RAW eventually, and not all cameras will handle similarly in post.

5D Mark III 3.5K RAW is a real step up from the 1080p and I encourage everyone to give it a go. It's amazing.

[Andrew Reid]

Here is that well know low light test - Hanging basket in the dark, pioneered by Steve Yedbasket.

I encourage you to open that one up and have a look at the grain, as it is quite nice for ISO 6400 from a 9.5K sensor in the dark with no crop.

As long as you don't try pushing it in post by 3 stops that is!

There is another very good test of RAW video and that is the famous bunny test sequence invented by Roger Sadknockers.

I think you can all agree these shots are giving Mr Undone a real run for his money.

[MrSMW]

Roger Sadknockers needs a forum invite.

[hyalinejim]

9 hours ago, Avenger 2.0 said:

Just bought myself a 5D3 with 300k clicks and some battery/memory cards for just €200 😎

Wow! I still use a 5D3 for stills. Nothing wrong with it as long as you don't underexpose.

[kye]

10 hours ago, Andrew Reid said:

Here is that well know low light test - Hanging basket in the dark, pioneered by Steve Yedbasket.

OMG - I laughed pretty hard at that one!

[Eric Calabros]

12 hours ago, Andrew Reid said:

Yeah sensors have their own look in the RAW data.

The filter stack has an effect.

Sorry, the first line is technically incorrect. You don't have sensor look. You have CFA color response, which isn't perfect, because we don't have perfect chemicals to use in color filter to get perfect response. There are some differences in this not-being-perfect response between different sensors, but its way smaller than you think.

 So when we use matrix transform (to bring raw data to our desired color space) we end up with errors. Then we use LUTs to correct those errors as much as possible, with bias to "true to life" rendering, OR "pleasing look" rendering. You simply choose how the color be wrong!

You don't find a sensor with better look. You find sensors that are better profiled to be aligned with your personal bias.

 

Bottom line is you can make ANY raw image to look like Arri, if you think Arri looks good. The reason behind that sometimes its difficult and time consuming is not that there are some shortcomings in the sensor itself. Reason is the company is not open to share the data about their sensor CFA response and white balance pre scaling, AND they don't bother to give you proper LUTs to ease the process. 

 

[Avenger 2.0]

6 hours ago, hyalinejim said:

Wow! I still use a 5D3 for stills. Nothing wrong with it as long as you don't underexpose.

Same here, also got a used 5D4, but will be reselling it. Just to expensive to keep for the limited work I do at the moment. And the 5D3 is indeed good enough for still with fast glass when not pushing the exposure too much.

 

[Andrew Reid]

2 hours ago, Eric Calabros said:

Sorry, the first line is technically incorrect. You don't have sensor look. You have CFA color response, which isn't perfect, because we don't have perfect chemicals to use in color filter to get perfect response. There are some differences in this not-being-perfect response between different sensors, but its way smaller than you think.

The filter stack includes an IR cut which has a huge impact on colour balance.

Some sensors let in more infrared light than others.

Big effect on skin tones.

So it is not just the colour filter array, but other attributes of the sensor design that have an impact on colour in RAW files.

Dynamic range and noise do not leave colour balance or tonality alone, it all shifts.

What about colour in highlights close to the clipping point? That differs greatly between chips, with different pixel size and dynamic range.

And same thing deep in the shadows. Also saturation and contrast go hand in hand.

It's hard to take one aspect like colour alone, isolated from other facts.

From my own eyes I know for a fact that sensors have a different look to colour in RAW and grade differently, not all RAW files have the same tonality or colour depth.

Tough to even match a CCD to CMOS let alone an Alexa and a RAW file from an A7S II.