Sigma Fp review and interview / Cinema DNG RAW

[Brian Williams]

4 hours ago, rawshooter said:

Would be really grateful if people who own/used the camera could clarify whether or not it crops the sensor when shooting 4K CinemaDNG.

Normally, RAW video means 1:1 sensor readout. Since the FP has a 6K sensor, this would then mean a crop from full frame (36mm) to APS-C (24mm). However, it seems - from reading the manual - that the APS-C crop is only optional and the camera can record 4K CinemaDNG with full sensor readout. Is it then binned RAW?!?

I'm not sure how to test other than by doing this, but here are two DNG's,  the first is a frame from CINE mode, a 4K DNG sequence (A001_002_20191105_000001.DNG), the second was taken in STILLS mode as a single DNG (SDIM0188.DNG).

Obviously the frame from STILLS mode is larger, but if you scale up the CDNG file, you'll see the framing matches, so no cropping (other than the vertical cropping to make it 16:9).

 

A001_002_20191105_000001.DNG SDIM0188.DNG

[rawshooter]

@Brian Williams - thank you so much for this! This clarifies a lot.

I'm still wondering how, technically, the camera downsamples the sensor's 6K Bayer pattern to the 4K Bayer pattern of the CinemaDNG frames...

[Lars Steenhoff]

1 hour ago, Brian Williams said:

I'm not sure how to test other than by doing this, but here are two DNG's,  the first is a frame from CINE mode, a 4K DNG sequence (A001_002_20191105_000001.DNG), the second was taken in STILLS mode as a single DNG (SDIM0188.DNG).

Obviously the frame from STILLS mode is larger, but if you scale up the CDNG file, you'll see the framing matches, so no cropping (other than the vertical cropping to make it 16:9).

 

A001_002_20191105_000001.DNG 8.07 MB · 1 download SDIM0188.DNG 33.72 MB · 1 download

Can you also make two cdng's one from the cropped apc mode and one full frame? 

[Brian Williams]

6 hours ago, Lars Steenhoff said:

Can you also make two cdng's one from the cropped apc mode and one full frame? 

Ok I’ll do that in a bit.

[paulinventome]

48 minutes ago, Brian Williams said:

Ok I’ll do that in a bit.

Thanks for the DNGs, it's really useful to compare the same frame in different modes. The two you did show that it is scaling the full width of the sensor and that the shadows in the 8 bit version are green. But overall a really interesting comparison. It would be super useful to do that with something that has dark to clipping in the scene and 8, 10 and 12 bit with a 14 bit stills control... But yeah, a bunch of work that you may have better things to do with your time! Cheers, Paul

[Llaasseerr]

On 11/5/2019 at 6:42 AM, cpc said:

 

You can use the gain control in linear spaces. With linear material gain is equivalent to exposure adjustment (although you don't get the intuitive numerical interpretability of a raw processing exposure control).

Exactly - it's not intuitive, or consistent across different working scenarios. I do use that. There is also the option to do it in the Offset tab in log space, but that is not particularly transparent either. In theory the more recently added right click option to bracket a grade operation within a certain color space should make this work a little better, but in practice I've found it doesn't work consistently and explicitly as it does in say, Nuke. The presumption in the design of Resolve that one would only want to do a linear exposure adjustment on raw footage in the raw tab is a little odd.

[rawshooter]

Thanks to @Brian Williams' two files, we can almost safely conclude that the Sigma FP skips sensor pixels when it saves the 6K sensor data into 4K CinemaDNG.

Here's a comparison between a 1:1 crop of

(a) the 6K still DNG developed with RawTherapee (with noise filtering, sharpening and other enhancements completely off) as a 16bit TIFF which was then downscaled to 4K using bicubic scaling (binning);

(b) the 4K CinemaDNG frame developed with RawTherapee without scaling.

 

 

The downscaled 6K DNG still (left) clearly shows more details and less noise than the 4K CinemaDNG (right). 

So effectively, the camera doesn't really benefit from the full frame sensor in CinemaDNG and might even produce a better video image in APS-C mode.

 

[Brian Williams]

10 hours ago, Lars Steenhoff said:

Can you also make two cdng's one from the cropped apc mode and one full frame? 

Ok here are two shots, both CINE mode, one uncropped, one APS-C cropped.

aps_crop_A001_005_20191106_000012.DNG non_crop_A001_006_20191106_000001.DNG

3 hours ago, paulinventome said:

Thanks for the DNGs, it's really useful to compare the same frame in different modes. The two you did show that it is scaling the full width of the sensor and that the shadows in the 8 bit version are green. But overall a really interesting comparison. It would be super useful to do that with something that has dark to clipping in the scene and 8, 10 and 12 bit with a 14 bit stills control... But yeah, a bunch of work that you may have better things to do with your time! Cheers, Paul

OK, here you go, 8bit , 10bit, 12bit all from CINE mode and a 14bit still- sorry it wasn't on a tripod so the camera moved slightly

still-SDIM0189.DNG 12bit-A001_007_20191106_000006.DNG 10bit-A001_006_20191106_000005.DNG 8bit-A001_005_20191106_000004.DNG

 

We really need that firmware update for shooting to SSD- I exchanged my 500GB T5 for the 1GB version in hopes it might fix the issue, but no, a handful of CDNG clips are still coming in with frames misplaced in the sequence, and I don't have the patience anymore to go through them and try to correct the order.

[rawshooter]

22 minutes ago, Brian Williams said:

Ok here are two shots, both CINE mode, one uncropped, one APS-C cropped.

aps_crop_A001_005_20191106_000012.DNG 8.06 MB · 0 downloads non_crop_A001_006_20191106_000001.DNG 8.06 MB · 0 downloads

1:1 crops of the above DNGs, developed with RawTherapee, without noise filtering/image improvements:

 
APS-C crop vs. full frame

 
APS-C crop vs. full frame

I'd say that APS-C is noisier...

So maybe I need to revise my earlier theory, and some binning is going on in full frame 4K (although it's not as good as the bicubic scale of a 6K DNG still to 4K with software in post); if the camera would just skip pixels, noise levels should have been identical.

[rawshooter]

1 hour ago, Brian Williams said:

OK, here you go, 8bit , 10bit, 12bit all from CINE mode and a 14bit still- sorry it wasn't on a tripod so the camera moved slightly

still-SDIM0189.DNG 30.73 MB · 0 downloads 12bit-A001_007_20191106_000006.DNG 12.05 MB · 0 downloads 10bit-A001_006_20191106_000005.DNG 10.05 MB · 0 downloads 8bit-A001_005_20191106_000004.DNG 8.06 MB · 0 downloads

Could it be that you mixed up the 8bit and the 10bit files?

After doing a bit of an extreme grading (Kodachrome LUT + pushing contrasts), I'm getting this:

12bit

10bit (really 8bit?)

8bit (really 10bit?)

12bit

10bit (really 8bit?)

8bit (really 10bit?)

 

[Brian Williams]

24 minutes ago, rawshooter said:

Could it be that you mixed up the 8bit and the 10bit files?

 

No, I'm pretty positive that's correct, especially looking at the sequential numbers of the file names, I definitely went up the line, 8-bit, 10, 12 in order.

EDIT- Yeah, I checked via raw digger, they are correct.

[rawshooter]

15 minutes ago, Brian Williams said:

No, I'm pretty positive that's correct, especially looking at the sequential numbers of the file names, I definitely went up the line, 8-bit, 10, 12 in order. Doesn't Rawdigger, or whatever that app is called, tell you the bit depth for each file?

You are right. I used Imagemagick's function for determining the number of discrete colors in an image, and the results were:

8bit-A001_005_20191106_000004.DNG: 3707837
10bit-A001_006_20191106_000005.DNG: 4068251
12bit-A001_007_20191106_000006.DNG: 5776266

The relatively little difference in total colors is probably related to the scene not having much contrast.

[cpc]

1 hour ago, rawshooter said:

So maybe I need to revise my earlier theory, and some binning is going on in full frame 4K (although it's not as good as the bicubic scale of a 6K DNG still to 4K with software in post); if the camera would just skip pixels, noise levels should have been identical.

You are comparing 6K Bayer-to-4K Bayer downscale + 4K debayer to 6K debayer + 6K RGB-to-4K RGB downscale. The first will never look as good as the second.

48 minutes ago, rawshooter said:

Could it be that you mixed up the 8bit and the 10bit files?

The 10-bit file is linear and the 8-bit file is non-linear. That's why 10-bit looks suspicious to you, it has lost a lot of precision in the darks.

4 hours ago, Llaasseerr said:

Exactly - it's not intuitive, or consistent across different working scenarios. I do use that. There is also the option to do it in the Offset tab in log space, but that is not particularly transparent either.

Yeah, well, the main difference with offsetting in log is that you are moving your "zero" around (a "log" curve is never a direct log conversion in the blacks), so you'd need to readjust the black point. Whereas with multiplication (gain) in linear there is no such problem. Still, offsetting is handy with log footage or film scans.

[Brian Williams]

On 10/18/2019 at 12:20 PM, cpc said:

It is not linear. I haven't looked at the exact curve, but it does non-linear companding for the 8-bit raw. The 12-bit image is linear.

 

On 10/19/2019 at 3:17 AM, paulinventome said:

Yes, i delved a little deeper and the 8bit DNGs have a linearisation table attached to them.

 

52 minutes ago, cpc said:

The 10-bit file is linear and the 8-bit file is non-linear.

I'm confused, it the 8-bit linear or not?

[cpc]

24 minutes ago, Brian Williams said:

I'm confused, it the 8-bit linear or not?

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

[Brian Williams]

11 minutes ago, cpc said:

12-bit is linear. 10-bit is linear. 8-bit is non-linear.

So when @paulinventome says "the 8bit DNGs have a linearisation table attached to them", am I misunderstanding what that means?

[cpc]

Just now, Brian Williams said:

So when @paulinventome says "the 8bit DNGs have a linearisation table attached to them", am I misunderstanding what that means?

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.

[paulinventome]

10 hours ago, Brian Williams said:

 

 

I'm confused, it the 8-bit linear or not?

Thanks so much, others have dived into this earlier and i'm still looking.

@rawdigger The 8 bit file is not linear, each DNG has a linearisation table which is the correct thing to do. I wanted to find out whether the 10 bit did the same and it seems that it might not. This means ironically that the 8 bit files in day to day use are probably going to to better quality (make better use of a limited format) than the 10 bit. As we know in linear the very last stop of light uses half the available values and the first 3 stops are represented by just 7 different values...

If there is one thing we should do - it is to petition sigma to encode the 10 bit files with a table like the 8. That way we can achieve the same sort of quality as the 12bit files.

Of course this is assuming they are not reading the sensor at this point in 10 bit mode which is one of the sensor modes possible.

IMHO this is a show stopper for me getting one as a B cam. 12 bit is only 24p and i need 25 at least.

I am waiting for clarification from sigma themselves but if others can chime in then maybe we can get them to consider doing this, i can't imagine it is *that* much work.

As for downsampling i am guessing they are scaling the RGGB layers before bayering it. I think by doing this they are making up for a lack of OLPF which is why the detail is lower.

I am so close to getting one as it's really what i am looking for, a sensor in a box that gives me pretty much what it sees. But there are a few caveats to sort through!

cheers
Paul

[MikhailA]

Guys , can anyone test moire/aliasing please?  Some test shots show it could be the big problem with FP .Besides, maybe this indirectly clarify  a down sampling algorithm...

[rawshooter]

A firmware update is out that fixes the CinemaDNG problems:

https://www.sigma-global.com/en/news/2019/11/08/2329/