IR has 5DIV studio samples and raws

Happy analyzing ...

http://www.imaging-resource.com/PRODS/canon-5d-iv/canon-5d-ivA7.HTM

br

me

First quick comparison vs 5D mkIII samples leaves me unimpressed. Noise levels not that good imho.

Beware, at the time of writing this reply, 5D4 samples are incorrectly hyperlinked from ISO200 and onwards, so to say, to see an ISO6400 sample, you have to clic on the ISO3200 link thumbnail.

What's your take on these, guys, regarding resolution and noise levels?

Cheers!

Not as good as the 1DX2 but not too bad. 12,800 looks usable

Guessing from jpg, more sharpness might mean less AA in low iso and less noise in high iso comparing 5D3.

I've never used Raw Therapee before, however upon downloading it and opening a few 5D Mark III RAW files, and then opening one of the Imaging Resource 5D Mark IV RAW files, it doesn't appear to work properly.

Yes, it'll open the file, but it has REALLY weird properties to it. This makes sense of course, because Canon has new RAW file data (the ability to do DPRAW).  Interestingly, it shows what I'm guessing are the masked pixels on the left side and upper part of the image.

I tried to DL the Canon Digital Photo Professional software linked to the 5D Mark IV on Canon's website, but they haven't even published it yet.

My version of DPP4 won't open the file and neither will LR.  So it appears, for the time being, that we'll have to wait on people to get the camera in-hand along with the latest release of DPP to check out the RAW files.

What do you use to view RAW of 5D4?

SidePod wrote:

Happy analyzing ...

Thanks for the notice. I checked sometime yesterday but no RAWs were there yet.

Anyway, in the ISO 100 "NRO" shot, the noise in the black borders is 2.5 ADU/DN, mean is 512.05, and the clipping point is 16383, so the ratio of clipping above black to noise is 6384, or 12.63 stops, at the pixel level. Normalizing to 8MP a la DxOMark's "Print", it comes out at about 13.6 stops.

That said, the noise may not be quite as random as what we have seen on some other cameras, like the 7D2, but I'm just judging from memory here. Even in the narrow black strip at the left edge of the image, I can see some banding, and it looks like Canon is using some scaling of individual color channels, because when you separate them, they have different noises and means, and the histogram from 500 to 524 looks a little jagged for the mosaic, and develops total gaps (like no 512s in one channel) when the individual channels' histograms are viewed. Well, I guess gaps are better than spikes that turn 2 values into one. Anyone interested in pure data can remove gaps algorithmically; they can't remove spikes without adding noise.

I don't see any banding in a strong push of the darker fabric on the upper left, though, so the banding is not visible for most practical pushing of shadows. It would probably take a more extreme scenario, like someone pushing ISO 100 to ISO 10K for the extended highlights for the banding to become an issue.

I can't see the black mug, because the program I used does not load the whole RAW; just the left half. I will try another approach later and see if I get the whole image. Being smooth, the black mug is a good area for heavy pushing to expose noise.

John Sheehy wrote:

SidePod wrote:

Happy analyzing ...

Thanks for the notice. I checked sometime yesterday but no RAWs were there yet.

Anyway, in the ISO 100 "NRO" shot, the noise in the black borders is 2.5 ADU/DN, mean is 512.05, and the clipping point is 16383, so the ratio of clipping above black to noise is 6384, or 12.63 stops, at the pixel level. Normalizing to 8MP a la DxOMark's "Print", it comes out at about 13.6 stops.

That said, the noise may not be quite as random as what we have seen on some other cameras, like the 7D2, but I'm just judging from memory here. Even in the narrow black strip at the left edge of the image, I can see some banding, and it looks like Canon is using some scaling of individual color channels, because when you separate them, they have different noises and means, and the histogram from 500 to 524 looks a little jagged for the mosaic, and develops total gaps (like no 512s in one channel) when the individual channels' histograms are viewed. Well, I guess gaps are better than spikes that turn 2 values into one. Anyone interested in pure data can remove gaps algorithmically; they can't remove spikes without adding noise.

I don't see any banding in a strong push of the darker fabric on the upper left, though, so the banding is not visible for most practical pushing of shadows. It would probably take a more extreme scenario, like someone pushing ISO 100 to ISO 10K for the extended highlights for the banding to become an issue.

I can't see the black mug, because the program I used does not load the whole RAW; just the left half. I will try another approach later and see if I get the whole image. Being smooth, the black mug is a good area for heavy pushing to expose noise.

Thanks for that analysis, John! My guess was about a 2 stop improvement over the 5D Mark II and III as DxO measures it, looks like I was off by 1/3 - 1/10 stop. Not bad given I didn't consult a calculator

It also appears based on your analysis (if I'm reading it right) that the character of the noise will be greatly improved (as we've seen with all higher end releases from the 6D onward) leading to perhaps another stop of IQ in terms of aesthetics/usability, depending on the viewer. Would that be fair to say or do you feel it's more user-dependent?

Anyway, in the ISO 100 "NRO" shot, the noise in the black borders is 2.5 ADU/DN, mean is 512.05, and the clipping point is 16383, so the ratio of clipping above black to noise is 6384, or 12.63 stops, at the pixel level. Normalizing to 8MP a la DxOMark's "Print", it comes out at about 13.6 stops.

Recommended clipping point is in tag 0x030f, it is 14008.

Yes, it'll open the file, but it has REALLY weird properties to it. This makes sense of course, because Canon has new RAW file data (the ability to do DPRAW).

Actually, it does not make much sense, the format is the same.

tonidavid wrote:

First quick comparison vs 5D mkIII samples leaves me unimpressed. Noise levels not that good imho.

It looks to me that raw files should have been exposed hotter. Underexposure leads to excessive noise.

Iliah Borg wrote:

Yes, it'll open the file, but it has REALLY weird properties to it. This makes sense of course, because Canon has new RAW file data (the ability to do DPRAW).

Actually, it does not make much sense, the format is the same.

Any chance you're seeing what I'm seeing with Raw Therapee? It's the image but it looks like a blown out, pink hued version of it.

It's the image but it looks like a blown out, pink hued version of it.

Tell them to use Canon black level tag from Makernotes We published it through exiftool loong ago.

Each time you see magenta fog and low contrast, it is the wrong black level.

Iliah Borg wrote:

It's the image but it looks like a blown out, pink hued version of it.

Tell them to use Canon black level tag from Makernotes We published it through exiftool loong ago.

Each time you see magenta fog and low contrast, it is the wrong black level.

AHH! Cool. Thanks! Never used Rae Therapee before, I assume I can set the black point. I will have to reinstall it and check when I get home.

Jonathan Brady wrote:

Iliah Borg wrote:

It's the image but it looks like a blown out, pink hued version of it.

Tell them to use Canon black level tag from Makernotes We published it through exiftool loong ago.

Each time you see magenta fog and low contrast, it is the wrong black level.

AHH! Cool. Thanks! Never used Rae Therapee before, I assume I can set the black point. I will have to reinstall it and check when I get home.

Umm. By "them" I mean "developers". The strategy is simple, for Canon raw from the cameras where they do not know the geometry of the optical black yet, to calculate black level directly from it, they can use the black level tag.

Iliah Borg wrote:

Anyway, in the ISO 100 "NRO" shot, the noise in the black borders is 2.5 ADU/DN, mean is 512.05, and the clipping point is 16383, so the ratio of clipping above black to noise is 6384, or 12.63 stops, at the pixel level. Normalizing to 8MP a la DxOMark's "Print", it comes out at about 13.6 stops.

Recommended clipping point is in tag 0x030f, it is 14008.

Why is it recommended? For standardization with in-between ISOs? Because it is non-linear above that point? Something else?

John Sheehy wrote:

Iliah Borg wrote:

Anyway, in the ISO 100 "NRO" shot, the noise in the black borders is 2.5 ADU/DN, mean is 512.05, and the clipping point is 16383, so the ratio of clipping above black to noise is 6384, or 12.63 stops, at the pixel level. Normalizing to 8MP a la DxOMark's "Print", it comes out at about 13.6 stops.

Recommended clipping point is in tag 0x030f, it is 14008.

Why is it recommended? For standardization with in-between ISOs? Because it is non-linear above that point? Something else?

It is calibration point, with data numbers above that one can get into the problem of "magenta highlights", and yes, it is linearity limit. One can say the value is rather conservative, but we are talking of 300-500 data numbers here that can be added without the risk of visible artifacts. In production body, Canon may not even allow the full scale of data numbers in raw, same as they do it on other bodies.

I expect we will see worthwhile improvement in the 5dIV images over the 5dIII.

One of the nice things with 5DIV is that it allows for 1 stop more headroom in highlights in dual pixel mode. That's actually a serious thing.