p.3 #1 · Canon 5D4 dynamic range analyzed from RAWs
Hathaway wrote:
For me I am hoping for the following real world improvement over the 5DSR. Increased ability to shoot at higher ISOs without as much of a noise penalty. Above ISO 1600 I start to get a lot of noise with the 5DSR that is difficult to deal with or causes the images to be unusable. This is primarily for wildlife. 5DSR is still a beast for landscape even though more DR would help in that situation you can always bracket for more DR.
In addition, it will help with harsh light situations for wildlife. These conditions create lots of shadows on birds, etc. that have to be lifted and start to introduce noise. I will be happy if there is more DR and less of a noise penalty for my shots.
I am really interested in comparisons between the 5DMIV and the 1DXII. I know I am getting extra MP in exchange for lower FPS, but I am wondering how they perform for ISO, DR and noise. ...Show more →
Not disagreeing that the iso performance of the 5dsr diminishes with iso but if you get the sun behind (even if behind a cloud or slightly down) you, I have been happy with wildlife to iso6400. And it is as good as the d810.
Performance of 5dsr at >=iso1600 is only exceeded by 1dxii .
1dxii cannot be printed as big because low mpx. 5dsr allows cropping that is not possible with 1dxii. 1dxii only works better if you can fill the frame or you require high fps and fast focusing (and can fill the frame significantly) .
p.3 #2 · Canon 5D4 dynamic range analyzed from RAWs
It seems that the 5D4 is a bit more complex than expected. I readed that the RGB channels had different read noise characteristics, and this is indeed true.
Here I compare two pictures from Imaging Resource, both at ISO 100:
1DX2 (image E1DX2hSLI000100NR0.CR2)
Left masked area of 72 pixels (using 68 at x=2, y=2)
red channel read noise: 2.30842 (DR at pixel level = 12.7472)
green1 channel read noise: 2.14933 (DR at pixel level = 12.8502)
green2 channel read noise: 2.18995 (DR at pixel level = 12.8232)
blue channel read noise: 2.21351 (DR at pixel level = 12.8078)
overall read noise: 2.21752 (DR at pixel level = 12.8052)
5D4 (image E5D4hSLI000100NR0.CR2)
left masked area of 136 pixels (using 132 at x=2, y=2)
red channel read noise: 2.10102 (DR at pixel level = 12.883)
green1 channel read noise: 2.48969 (DR at pixel level = 12.6381)
green2 channel read noise: 2.44241 (DR at pixel level = 12.6658)
blue channel read noise: 2.86546 (DR at pixel level = 12.4353)
overall: 2.49716 (DR at pixel level = 12.6338)
The 1DX2 seems to have about the same read noise in all channels (as all previous Canon cameras I suspect). But this is not the case in the 5D4. This confirmed with 2 different images. The red channel read noise is about 0.85x that of the greens, and the blue channel is about 1.15x the greens.
This makes the red channel the one with less noise and indirectly more DR (theoretical, because any white balance will destroy it). Perhaps Canon is trying to improve the red shadows as much as possible, calibrating the sensor DR peak for it.
With these misteries I had no time to get the proper comparison between both cameras. At least it is now clear that 5D4 is about 13.6 EV of DR, after looking several ISO 100 RAWs it seems that the more common read noise is around 2.5 (bellow the 3 in my first images from dpreview).
Several points I can see;
1) At native resolution the 5DIV is competitive with the A7RII at +5EV and equalised only slightly worse, less than 0.5EV IMO.
2) At native resolution the 5DIV is about the same as the 1DXII and equalised slightly better.
3) The 5DsR effective DR is better than what many claim, as the shadow noise and lack of banding is much improved over 5D3 and even 6D. 5DsR is competitive to +4EV, while not as clean even equalised it still pretty good and cleans up well in post processing. I'd probably keep it to +3.5EV max, but for 5D3 I'd only do +2EV.
4) D810 is king at ISO 64, but at ISO 100 Pentax K-1 is new king!
In other news Canon is working on next gen lithography, maybe this is only beginning of improvements.
Interesting comment about the red channel, because it was clear in dpreview studio scene 5DIV was not oversaturating reds which it is famous for, anymore. I wasn't sure if it was due to beta nature of RAW converter as other colours were a bit muted or a real improvement.
p.3 #5 · Canon 5D4 dynamic range analyzed from RAWs
skibum5 wrote:
EDIT: I do wonder though if it might end up only grabbing light from half the incoming angles though for that top stop and might that not cause some issues?
At the least we would end with less sharpness (half the resolution) if we only use the secondary subframe.
We can safely assume that there is zero highlight recovery capability from the secondary subframe. When Canon combines the two subpixels (it seems that 14 EV each) they yield a 15 EV image. Yes, they need to discard 1 EV, but obviously the logical choice is to discard 1 EV from the shadows (mostly noise) and not from the highlights.
The secondary subframe could only provide a few additional unclipped pixels (0.0001 EV?) sacrifized during the rounding manipulations combining the images, but on no way a full stop of DR.
Yes, some RAW manipulations are not intelligent (e.g. in ISOs above 6400, instead of storing a tag signaling the boost, the physical values are scaled clipping the highlights). Likely for convenience with the converters.
p.3 #6 · Canon 5D4 dynamic range analyzed from RAWs
Pixel Perfect wrote:
My summary I posted at dpreview
Several points I can see;
1) At native resolution the 5DIV is competitive with the A7RII at +5EV and equalised only slightly worse, less than 0.5EV IMO.
hmm 5D4 looks almost nearing more like a stop worse than A7R II to me, set to +6 ISO 100 for everything and move the window around to some darker parts of the frame than the way too bright default position, seems like rather noticeable difference to me, I fear being that far back, would be the sort of thing where the improvement is very nice and yet it will still leave you a bit wanting for some single shot real world high DR stuff, since you can only just manage most of that single shot with a7R II, you could sort of pull it off but will definitely get some uglies
I do note that if the supposed dual raw magic 1 extra stop actually somehow is true that the 5D4 would suddenly be right up there 100.000% though....
p.3 #7 · Canon 5D4 dynamic range analyzed from RAWs
cgarcia wrote:
At the least we would end with less sharpness (half the resolution) if we only use the secondary subframe.
We can safely assume that there is zero highlight recovery capability from the secondary subframe. When Canon combines the two subpixels (it seems that 14 EV each) they yield a 15 EV image. Yes, they need to discard 1 EV, but obviously the logical choice is to discard 1 EV from the shadows (mostly noise) and not from the highlights.
The secondary subframe could only provide a few additional unclipped pixels (0.0001 EV?) sacrifized during the rounding manipulations combining the images, but on no way a full stop of DR.
Yes, some RAW manipulations are not intelligent (e.g. in ISOs above 6400, instead of storing a tag signaling the boost, the physical values are scaled clipping the highlights). Likely for convenience with the converters.
Manufacturers are lazy, but no darn stupid... ...Show more →
yeah it does seem weird, i'm not sure why and how dual pixel raw should be able to retrieve an extra stop since it's not like they are running each half at a different ISO and why would they under-utilize the well capacity and so on as default and not have thought of this it if was possible, etc.
so i am becoming more dubious
that said Iliah Borg often knows what he is talking about, so maybe his magic is not magic but real, becomign a little dubious, but it would be fantastic if true, he seems to claim the full RAW show solid clipped colors in highlights where he claims you can see well distinguished values and he claims there seems to be no weird histograms holes or anything, but I agree it does seem all a bit strange and seems a little hard to believe....
p.3 #8 · Canon 5D4 dynamic range analyzed from RAWs
Pixel Perfect wrote:
My summary I posted at dpreview
Several points I can see;
1) At native resolution the 5DIV is competitive with the A7RII at +5EV and equalised only slightly worse, less than 0.5EV IMO.
2) At native resolution the 5DIV is about the same as the 1DXII and equalised slightly better.
3) The 5DsR effective DR is better than what many claim, as the shadow noise and lack of banding is much improved over 5D3 and even 6D. 5DsR is competitive to +4EV, while not as clean even equalised it still pretty good and cleans up well in post processing. I'd probably keep it to +3.5EV max, but for 5D3 I'd only do +2EV.
4) D810 is king at ISO 64, but at ISO 100 Pentax K-1 is new king!
In other news Canon is working on next gen lithography, maybe this is only beginning of improvements.
Interesting comment about the red channel, because it was clear in dpreview studio scene 5DIV was not oversaturating reds which it is famous for, anymore. I wasn't sure if it was due to beat nature of RAW converter as other colours were a bit muted or a real improvement. ...Show more →
IQ results look very good. The only issue now is possible lack of sharpness.
p.3 #9 · Canon 5D4 dynamic range analyzed from RAWs
skibum5 wrote:
hmm 5D4 looks almost nearing more like a stop worse than A7R II to me, set to +6 ISO 100 for everything and move the window around to some darker parts of the frame than the way too bright default position, seems like rather noticeable difference to me, I fear being that far back, would be the sort of thing where the improvement is very nice and yet it will still leave you a bit wanting for some single shot real world high DR stuff, since you can only just manage most of that single shot with a7R II, you could sort of pull it off but will definitely get some uglies
I do note that if the supposed dual raw magic 1 extra stop actually somehow is true that the 5D4 would suddenly be right up there 100.000% though.... ...Show more →
I said at +5EV, I agree above that it's about a stop worse, but I wouldn't go above +5EV and that's a huge boost for most work. Canon has improved shadow DR by at least 3EV since 5D3 came out, so that's is good enough for me. I'm already finding 5DsR a good improvement, and this is even better.
p.3 #10 · Canon 5D4 dynamic range analyzed from RAWs
EB-1 wrote:
IQ results look very good. The only issue now is possible lack of sharpness.
EBH
ACR is not the best at detail extraction, but this is beta software. DPP will do better, but hopefully so will DxO, C1 and On 1 if they ever ship their RAW converter.
p.3 #13 · Canon 5D4 dynamic range analyzed from RAWs
Pixel Perfect wrote:
Well then I expect to see better results soon.
I wish Adobe would allow ACR to have sharpening radius from 0.1 rather than 0.5, cameras like the 5DsR need a lighter touch.
?? ACR can set anything from 0.1 and up and then you can use stuff like NIK Sharpener later if need be.
Unless the new 'magical 'cloud'' i.e. rental model ACR has changed for the worse....
p.3 #15 · Canon 5D4 dynamic range analyzed from RAWs
The 5D4 sees a very significant increase in DR that is very, very welcome to people who care about such things. What you can say now is that it is at least in range, as opposed to woefully behind.
But the A7rII, and to an even greater extent the D810, do still show visible advantages even from +3 EV. Obviously they're not huge, and they never become huge (significant at +5-6 EV, as as those EVs grow, the advantages may get bigger, but the frequency of calling of them grows smaller).
The other issue is that the D810 (presumably the not-distance D820), and to an even greater extent A7rII (and Sony seems to move fast) have higher resolution, which impacts the equation. Given where we are in the upgrade cycles we can also assume that those competitors are far closer to making that gap bigger.
So for me, what it means is this: The real win on this (in terms of landscape) is when Canon brings this tech over to the 5Dsr line. It's virtually assured this will happen-- but the timing is important and unknown.
Really happy to see Canon make this big step ahead!
p.3 #16 · Canon 5D4 dynamic range analyzed from RAWs
skibum5 wrote:
?? ACR can set anything from 0.1 and up and then you can use stuff like NIK Sharpener later if need be.
Unless the new 'magical 'cloud'' i.e. rental model ACR has changed for the worse....
You can set radius in Unsharp Mask in Photoshop from 0.1, In ACR minimal value for radius is 0.5.
p.3 #18 · Canon 5D4 dynamic range analyzed from RAWs
skibum5 wrote:
?? ACR can set anything from 0.1 and up and then you can use stuff like NIK Sharpener later if need be.
Unless the new 'magical 'cloud'' i.e. rental model ACR has changed for the worse....
Not true, ACR has 0.5 minimum and has never been able to go lower, in fact 0.5 is the improved version, at once stage it used to be 0.7.
p.3 #19 · Canon 5D4 dynamic range analyzed from RAWs
skibum5 wrote:
apparently on DPR there is talk that DUalPixel RAW will somehow allow an extra stop of highlights though
cgarcia wrote:
Can't be true (outside technical stuff: if it were so easy to increase the DR by using smaller subpixels, all DSLRs long time ago would have been using phone-camera-sized pixels combined).
It can be true - you can go and buy a camera that does exactly that:
And no-one said it was easy, this method of increasing dynamic range has some disadvantages: it only works in good light whilst reducing sensitivity and DR in low light, it requires extra processing, and it doesn't make the sensor "ISO-less" like eliminating read noise does. That is why manufacturers have gone down the road of trying to get read noise as low as possible.
cgarcia wrote:
The ADCs are more than capable to extract all the truly available DR, and nobody (except maybe God) can change noise into signal...
You can increase DR without changing noise into signal. DR is the difference between the noise floor and the saturation point, raising the saturation point alone will increase DR for the same [absolute] level of noise in the image.
It depends exactly how the Canon system works. If both pixel halves are digitised to 14-bits, summed and then written back as a 14-bit value then yes you can gain DR (but less than a stop since you doing two reads hence your read noise increases - assuming the charge from the subpixels is combined before readout in non DPRAW mode).
