Lars Steenhoff

I have it and its much faster than the old version in resolve, I have realtime playback and the old version I did not.

I have it and its much faster than the old version in resolve, I have realtime playback and the old version I did not.

I have it and its much faster than the old version in resolve, I have realtime playback and the old version I did not.

I have it and its much faster than the old version in resolve, I have realtime playback and the old version I did not.

The charts and the sharpness tests are tough metrics and not real world either.
AFAIK the downsampling is in some way to minimise moire because there is no OLPF. You can't really have an OLPF in a camera that switches modes or does stills. I've seen plenty of shots ruined because of that from some quite well known 'prosumer' cameras as well. And BMD is quite notorious for this. They score well on tests but not so much in real world.
But the sigma image is quite soft. 
What the slashcam shows though is a difference between crop and normal and that's interesting. So full frame is downsampled but then it's also fair to say the downsampling isn't an even number and maybe that's why we're seeing these kind of results whereas the S35 mode might actually be putting the sensor into a windowed mode and that's worth looking at more.
The 12 bit footage - i strongly think if you are recording 12 bit linear to push the image as high as you can and bring down. Because that really is linear, the brightest stop of light will show nearly 2000 steps of gradation. So push as far as you can to take advantage of the tonality. Of course that's not always possible. The 8 bit footage less so because they are reshaping the data so that top stop takes much less space. Again i'd really hope for 10 shaped bit from 12 bit linear.
I do believe if this is the IMX sensor then there is a 4K 14bit crop mode which would be awesome.
But with no OLPF then as an overall format you'd see much more aliasing.
Lightroom is quite a bad debayer - i see artefacts all over the palace. Resolve is a bit better. SpeedGrade was poor. And nuke used to have some really good options in there (but those got 'upgraded')
Motion RAW is quite a bit more difficult than stills RAW and i don't think there is a good DNG option for debayering out there. It would be possible to use temporal debayer but i am no aware of anything like that which exists...
Also the reason why it seems better to push the shadows that bring the highlights down is something that just occurred to me on a long drive back. The 12 bit is just truncating the top stops. The whole camera exposure is the same for stills vs cine. So same settings like for like, but the cine has the 2 bits chopped off. So therefore the expectation is for there to be highlights compared to mid grey and it's not quite there. That 14 bit 4K mode would be perfect!
cheers
Paul

It has a cleaner interface and runs faster due to GPU acceleration. I would greatly recommend it over the older version. I don't use it all the time, but its great when I'm trying to bring up underexposed footage.  

Thank you.
I have looked all over YouTube for sample footage, and most of the results show videos that are overexposed, with highlights clipped, I guess they didn't shoot RAW or didn't have a way to verify the exposition. The best video available on YouTube is this, which has the RAW sample footage available for download: 
 
Pixel binning I guess, that was my concern, so, just like the Sony a7III, the Nikon z6, the EOS R, the DJI Mavic 2 Pro, etc. if you want sharper results, avoid downscaling, use crop mode :-/ 

Scene dependent really and what you're looking to achieve.
- Outside with clouds - you can set zebra to 100 and when the clouds clip you can recover some of that because the green and red channels won't have clipped (getting a RAW clipping indicator for each RGB channel would be really useful)
- If it's dark and you have latitude, better to expose up and bump up the ISO and then take that down to get cleaner shadows
- But if you're going to clip the highlights you really need, then bumping up the shadows will the best bet - as you say.
In reality it's probably safer in a full dynamic range scene to expose 'properly' and have a bit of latitude for moving stuff around.
Just be mindful that the shadows, at the sensor side, have progressively less tonality the lower you go and all sensors are like that!
cheers
Paul

AFAIK it's a mode on the sensor itself so the data coming off the sensor is binned in that mode. This is why the rolling shutter is so much better in that specific mode. It is also 12 bit even to SD Card.
However the FHD can show aliasing on smooth lines
IMHO you are better shooting UHD (even in 8 bit) and downsampling to HD in post. Much, much better result assuming you can achieve the frame rate you want.
The sensor (if it is the one we all think it is) does have a 4K DCI binned mode in 14 bit - this would be really nice to have. Otherwise all the cDNG formats are 12 bit max, so you will only see 12 stops max (real stops)  from the image. I have tested stills DNG with cine DNG and can quite clearly see the missing stops at the top end. I would say it's closer to 1 to 1.5 stops difference.
But....
If you look at the S1H for example (same sensor perhaps?) in 6K mode C5D is claiming 12.7 stops of range. Well the difference is that the S1H is *probably* doing highlight reconstruction as part of white balancing and baking out a movie. This would yield a few more stops on a wedge chart (it's ideal for wedge charts!). It would be interesting to see a RAW test with the S1H. But the fp will get you the same range, with reconstruction (Lightroom does it anyway and resolve it's a checkbox) perhaps more.
Colourwise it's good. I can match Red footage but the Red has more range, at least 2 or 3 stops more on a practical basis. I've shot side by side through ranges. The ISO matches fairly well as well but i would tend to expose the fp slightly higher when possible to protect the shadows.
As for those Videos - RAW is not simple to handle properly - simple as that.
The second video shows the guy really has *no idea* what he's talking about or understands what RAW is and sensors in general, it's almost comically bad.
The bad blacks - it's difficult to say without seeing the RAW he shot - i suspect it is massively underexposed and he's pulled the shadows up. The thing about RAW is that with no compression you will see each pixel the sensor saw. As there is no noise there i guess the ISO was very low therefore probably just failed to expose correctly. Because it's linear - as i keep saying - the first 3 stops are represented in 7 code values. Usually compression just screws this totally and they're expanded out and noise from compression hides this but this is what the sensor (any sensor!) sees
cheers
Paul

Very impressive at -3 stops, doesn't lose saturation like the Pocket 4K and X-T3 do.
 
This guy is not a fan. I suspect he's never graded linear raw before.
 

Yup.
Recording RAW is a pro choice, there's no way around that and for holiday videos and snaps as you've found, isn't ideal. Maybe once they include some kind of log recording through firmware it would be better suited to that. But i see this as a sensor in a box and you build everything around that.
The cinematic DNG test that Lars posted back there is perhaps some of the best footage that's been publicly shown from the camera. These are very challenging scenes for any camera and the fp does a fantastic job. However their post workflow isn't ideal - i see exposure wavering between frames which is either a factor of the firmware version they are using or the fact that appears to be batch processed through lightroom.
The native gamut of the camera is certainly beyond 709, so therefore as i guessed that workflow is going to be the nut to crack. I've shot primaries off a reference monitor in P3 and 709 and can clearly see the difference. So any workflow built around 709 (Adobe) will somehow be messing with the native gamut to squeeze it into the 709 bucket meaning colour clipping or scaling - all warning signs for me that the final image will be compromised. Another reason why the camera isn't for the masses.
I personally would not archive DNGs in any log or compressed format but slimRAW is a good choice. I don't see the point of having a camera that can dump out RAW then transcoding to a less format. Just get a camera that shoots in the format you need.
This will be sigmas marketing issue. I understand what they're aiming for but it's a small market. Steve Huff certainly has fallen for the stills side and i get my VF today i think, in which case then i can spend some time using it like that. 
cheers
Paul
 
 

You also cannot use the zoom while recording, I would love to have this, sigma learn from sony and fuji

A bit more info for those interested.
Firstly, as expected, setting white balance in camera does not affect the RAW file. There have been cameras that have baked white balance before, the Sony FS700 for one used to do that (or the FS7, one of them did). But that does not happen here.
Secondly i'm enclosing the image of the linearisation curve in 8 bit. It's a good curve, it's not log but redistributes the values in a way that makes sense.
Thirdly. Anyone worked out WTF is going on with focus magnification? So in cine mode, i can assign the AEL button to zoom. But how do i move around the screen to focus on different areas. I would have thought the control ring would (like on any other camera). It is possible - you can do it through elevntybillion key presses. Also pressing the touch screen would make sense but that just puts the auto focus points up... I must be missing something here!
cheers
Paul
 

I'm glad they're adding log, but I would only use it for monitoring raw recording. For that, it's crucial. The log image with LUT will then match the RAW image with LUT in Resolve (different input but same output image). 
So as long as:
- They publish the log curve and wide gamut they are using. They need a conventional Cineon/AlexaLogC/ACEScct type of log curve that will hold all the highlight information of the raw sensor.
- It can send a log signal over HDMI so that you can record DNG raw and monitor with a lut on the HDMI log image. If it's sending out raw over HDMI that's quite cool, but as far as monitoring,  you are limited to the Atomos Ninja V vs. every other EVF or external monitor on the market that allows a custom LUT. 
 
As for exposing raw, the only two things you need to think about are:
1. Expose with a grey card and a lightmeter like you are exposing film. You can also use false colour and a grey card but just make sure you're aware of what gamma setting the false colour is expecting, or make your own false colour LUT.
2. Check where the highlights are clipping. If you need to protect the highlights further, then underexpose by 1 or two stops and push by the same amount in post. Use Neat Video to clean up the noise floor if necessary.
Zebras are for a video world. They can be semi relevant for checking highlights in log since it puts everything in a 0-1 range. Again, we need log output for monitoring and if for example shooting for ACES we can put a log to ACES rrt/odt LUT on the camera monitor/EVF and see how the full dynamic range of the highlights are rolling off. Then get the exact same result on the ingested DNG raw images to ACES in Resolve.
 
Log for monitoring raw recording is crucial as it allows any Chinese 8 bit monitor with custom LUT option to display the full raw image dynamic range and get a very close match to what you will see in Resolve with your beautiful DNGs - as long as you have your technical LUTs set up correctly.
Published log and gamut is a MUST though. Sigma, please do a white paper documenting this. This way, we can do a direct correlation with the linear raw image. And allow sending log over HDMI while recording raw.

Wonder if this thing really is up to snuff: https://www.kickstarter.com/projects/2124504218/warpdrive-high-speed-portable-ssd/description

Yes, drop them a line and ask. That's what i mean. Doesn't have to be log, the values they choose can match the data better - it could be a log with an s curve to distribute mid tones better. We will never see that curve, we will only ever see the results after the curve.
Whilst the DAC is 14 bit i am pretty sure that the mode they are running the sensor in is 12 bit output for motion. And a 10 bit DNG with a linearisation table would hold everything a 12 bit linear file could - i see no reason for a 12 bit option at all.
I think 4096 across should be an option though, even if it's scope (4096 x 1716) or 2:1 (netflix style 4096 x 2048) - the overall data rate would be roughly the same as UHD.
 
I think a couple of wish list items. Yes, a user LUT to output baked formats would be useful in some cases.
But tools to expose for RAW better - i'm still trying to wrap my head around the zebra at 100% and what that actually means.
cheers
Paul

It would be nice to have LOG if you ever wanted to shoot non RAW for a project that didn't require it. 

I've been pushing and pulling the 8 bit DNGs in lightroom, comparing to 12 bit FHD and also Sony A7sII stills.
Whilst i can see the limits in the shadows, as there are no compression artefacts, then what you get is remarkably usable. Pushing something 5 stops over to compare shows the 8 bit DNGs really not that far from the Sony RAW and 12 bit versions. So i'm running to a 128GB SD Card (ProGrades) and have yet to really break the image.
This is less to do with the sensor and more to do with no compression. That's the key, compression kills everything.
So i don't know what you're planning on shooting at the moment i'm skewing towards 8 bit.
10 bit linear is no use whatsoever. 12 bit 24p would make a difference but i can't use 24p
As i mentioned above, if we users can drop sigma a line and show how much we want 10 bit DNG with a linearisation table then that would be the sweet spot and i would externally record. If we're lucky they could experiment with writing 10 bit to SD card as well... And maybe do some other frame sizes... (4096 DCI widescreen!)
cheers
Paul

I think you will find that the VF attachment will make a big difference.
 

I think you will find that the VF attachment will make a big difference.
 

I succumbed and bought one. 
I will test more fully over the next week. Ergonomically from a stills view it's very much a urgh. No EVF makes life very difficult for me but i do have the VF on order so maybe that will change things.
Cine mode. DNG. It's heaven. This is all you need from a camera. I've side by sided UHD 8bit onto SDXC vs A7sII UHD Slog2/Sgamut and there's no joke. It's almost comical. The low end and the highlights are literally playing in different ballparks. What crushes the A7sII footage is the compression. Simple as that. So that 8 bit UHD is remarkably good. I'm pushing and pulling it and comparing to the same thing shot  FHD in 12 bit and it's incredibly robust - so much more than it should be.
Over the next week i'll push and pull properly. I got it as I needed a lightweight B cam to my Red for a shoot - so i need to intercut and i'll be finding it's comfort zones side by side with everything.
So far the best resolve post in in ACES. My fear with going into any 709 space is that the camera native spectral response will be outside of 709. All Sony sensors push red way out and it's easy to see with bright red subjects. If you don't debayer into a larger colourspace then those reds get clipped or mushed into gamut. I don't know enough about DNG support in Resolve but visually into ACES appears much better. The only other valid option i think is Linear in, certainly none of the BMD Presets as they're for very different sensors.
 
Cheers
Paul

First I don't think this is true, and even If it is, then it also applies to a compressed raw in a still camera. 
When the next generation stills cameras from Nikon or canon can shoot stills at 24 frames per second with compressed raw red should start a court case against them.
 

The linearisation table is used by the raw processing software to invert the non-linear pixel values back to linear space. This is why you can have any non-linear curve applied to the raw values (with the purpose of sticking higher dynamic range into limited coding space), and your raw processor still won't get confused and will show the image properly. The actual raw processing happens after this linearisation.

12-bit is linear. 10-bit is linear. 8-bit is non-linear. No idea why Sigma didn't do 10-bit non linear, seeing as they already do it for 8-bit.
Here is how 10-bit non linear can look (made from your 12-bit linear sample with slimraw). In particular, note how darks are indistinguishable from the 12-bit original.
10-bit non linear (made from the 12-bit).DNG