They wouldn't be so stupid, but I haven't seen an announcement from Canon that unambiguously claims 8K DCI.
As in the comments above, the promise was '8K/4K DCI'.
Upvote
0
The Canon EOS R5 and EOS R6 will be announced in the first few days of July
- Thread starter Canon Rumors Guy
- Start date
They wouldn't be so stupid, but I haven't seen an announcement from Canon that unambiguously claims 8K DCI.
As in the comments above, the promise was '8K/4K DCI'.
| Camera Model | Low Light ISO |
|---|
| Canon EOS 1D X Mark II | 5189 |
| Canon EOS 1D X Mark III | 4915 |
| Canon EOS 5D Mark IV | 5011 |
Photographic Dynamic Range versus ISO Setting
www.photonstophotos.net
so.. IDxII somewhat better than 5DIV at low light ISO and 5dIV somewhat better than 1DxIII.
the difference is negligible though, as you can see.
Upvote
0
If Nelu can't tell the difference in high ISO performance between the sensor in the 5D Mark IV/EOS R and the sensor in the 1D X Mark III, you're probably asking the wrong person that question.
And the 5D Mark IV sensor when they put it in the EOS R...
And the 18 MP APS-C sensor when they put it in a gazillion different cameras...
Upvote
0
Meh. The difference between 8,208 x 5,472 (got to make both sides divisible by 16 for efficient JPEG compression) and 7,680 x 5,120 is only the difference between photosites 4.39µm and 4.69µm wide. That's less than 7% in terms of linear resolution.
Upvote
0
My guess is, because Canon claims 4K DCI uses the full width of the sensor, it makes 44.7mp more probablebut not certain. Due to Bayer filters used in Canon sensors, the easiest way (but not the best in terms of the image quality) to get both 8KDCI and 4K DCI will be to skip pixels by 2 over 2, like this:
0, 1, 4, 5, 8, 9, 12, 13, 16, 17...
There's also pixel binning technique where they would merge data from groups of pixels, it's also easier to do if the sensor has 8192px.
If that's the case, 4K DCI implies 8K DCI and therefore, 44.7mp pixel count.
While you're at it, make both sides of the 3:2 sensor divisible by 16 for efficient JPEG compression with fewer artifacts. So 8,208 x 5,472 = 44,914,176.
The video will use 8,196/8,208 = 99.85% of the sensor width, which is close enough to "full" when one must consider that there will also be several rows of pixels beyond 8,208 x5,472 on each edge that are needed for interpolating color information on the edges of the output, as well as a few rows of masked pixels.
Upvote
0
Not sure if Canon is really concerned about efficient jpeg compression when they're scratching the limits of the sensor readout speeds. Also they're now pushing HEIF format.
But as to the point of 39mp goodness, I have 5DIV with 30mp, and a prospective upgrade to a 39mp camera doesn't sound very exciting in terms of landscape photography. 45mp is probably right about where my threshold of GAS sits.
Upvote
0
That's a number based on an arbitrary "line" above which everything is acceptable and below which nothing is acceptable. As one can see from the actual graphs, exactly where one draws the arbitrary "line" determines which camera does slightly better or worse because the measured performance lines criss-cross each other multiple times between ISO 100 and ISO 102,400.
What that type of test never reveals is how different processing algorithms from one camera model to the next in raw processing applications that photographers actually use to process their photos can also affect what the final output looks like. Those tests can also be very sensitive to the differences in copy variation between two supposedly "identical" sensors with regard to how many defective pixels each one actually has (and they all have some - that's why pixel mapping is performed on every single camera at the factory), where on the sensor they are located, and what color filter is in front of each of them.
Last edited:
Upvote
0
Using blocks divisible by 16 greatly reduces the number of artifacts due to JPEG compression. For video this is not so crucial, but for stills it is. Canon has never released an EOS camera in any format that does not have a sensor with both sides divisible by 16. HEIF also uses compression. Due to the way HEIF compresses image data, using blocks divisible by 16, or at a bare minimum divisible by 8, is probably even more critical for HEIF than JPEG, but I've not looked at HEIF in detail.
Upvote
0
I think jpeg inefficiency would mean just marginally larger jpeg files (marfinally worse compression rate) which may not be a big deal.
Yes they'll also need a number of rows of shadowed pixels but I don't know if they include them in the official specs. For 8K DCI they need to have at least 8192 active pixels on the longest side, maybe more for jpeg compression as you said.
That really is insignificant; 44.7 or 44.9mp doesn't matter much to me. 39mp would be disappointing...
Upvote
0
Look also at shadow improvement and read noise. Shadow improvement shows that the 1D X Mark III is more ISO invariant than the previous models, and read noise should be self explanatory. Lower is better on both charts.
Last edited:
Upvote
0
I think jpeg inefficiency would mean just marginally larger jpeg files (marfinally worse compression rate) which may not be a big deal.
Yes they'll also need a number of rows of shadowed pixels but I don't know if they include them in the official specs. For 8K DCI they need to have at least 8192 active pixels on the longest side, maybe more for jpeg compression as you said.
That really is insignificant; 44.7 or 44.9mp doesn't matter much to me. 39mp would be disappointing...
Again, for whatever reasons, every EOS sensor Canon has ever produced has both sides divisible by 16 at maximum still image resolution. I doubt they'll stray from that any time soon.
Now let the "cripple hammer" cries begin because Canon promised "full width" and then delivered a 1.00146X crop factor for 8K video!
Upvote
0
For which I think they just add padding pixels to the edges. That has a minor impact on the compression rate.
Hmm that may be related to the technicalities of the sensor design, not jpeg compression. Although in this case your calculations of the R5 image size are still valid.
Upvote
0
I've suggested the same before, though I do not remember if it was on this thread or on others (probably both). That suggestion was roundly criticized, but I wouldn't be surprised. There does not seem to be much, if any, difference in the actual sensor hardware between the 1D Mark II and 1D X Mark III other than the different low-pass/anti-aliasing filter in front of them. What marginal differences there are seem to be more along the lines of how the analog signal is processed before it gets to the ADC. That and improved raw conversion algorithms, at least with Canon's JPEG engine in the camera and within their own Digital Photo Professional. What algorithms third party raw convertors use for any sensor is entirely up to the third party that produces the application.
Any EF mount sensor used in an RF mount body will get new microlens arrays in front of the actual sensor chip to account for the different angles of incidence on the edges and in the corners of light coming from the shorter registration distance for wider angle lenses. That's why the 5D Mark IV sensor used in the EOS R was a "new"sensor. Ditto for the 6D Mark II sensor used in the EOS RP.
Last edited:
Upvote
0
The hardware JPEG engine needs to be a lot more complex if you allow for images that don't have sides divisible by 16. And if you do a naive implementation to save memory, it will throw of memory and cache access making the while pipeline do the computer equivalent of stuttering.
And for people paying attention at home: 1080 isn't divisible by 16, ever wondered why there's a coloured line at the bottom of a video sometimes?
Upvote
0
It has nothing to do with compression rate. It has to do with compression efficiency, as in how many compression artifacts the results demonstrate. A more efficient compression process can produce smaller files with the same image quality as a less efficient compression process can produce making larger final files (assuming the scene content/raw image data is identical).
The reason they use output dimensions divisible by 16 is all about compression efficiency. There are tons of white papers and other scholarly resources that discuss this for compression in general, not just as it is implemented by Canon.
Upvote
0
Similar threads
- Contains 1 staff post(s)
