The Canon EOS R5 and EOS R6 will be announced in the first few days of July

Michael Clark

Now we see through a glass, darkly...
Apr 5, 2016
4,722
2,655
Adding padding fake pixels will make the h/w stutter?

They don't add any "fake pixels". They add actual hardware photosites to the sensor outside the edges of the output resolution because color interpolation is done using surrounding photosites, both those that are filtered by the same color as the photosite that corresponds to the pixel on the edge of the output image as well as those that are filtered by different colors than the photosite in question. If a demosaicing algorithm uses three rows of photosites in all four directions to interpolate colors for the output pixel representing a particular photosite, then the sensor needs three rows of additional photosites on all four sides of the sensor beyond the number of rows and columns output in the final image. If a demosaicing algorithm uses five rows of photosites in all four directions, then the sensor needs five rows of additional photosites on all four sides, and so on.

Demosaicing algorithms to do color interpolation have nothing to do with compression algorithms. Those are two entirely separate things in the processing pipeline.

After demosaicing (including color channel multiplication for color temperature and white balance correction, gamma processing/light curves are applied) is done, there is an uncompressed image file that has RGB values for every pixel in the output resolution. Uncompressed TIFF files are one form that represents these individual RGB values for each pixel. Then whatever image file compression scheme used is applied to the uncompressed RGB values for each image pixel.
 
Last edited:
Upvote 0

Michael Clark

Now we see through a glass, darkly...
Apr 5, 2016
4,722
2,655
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?


Also, if you are preparing an image for print and are cropping it, if you crop it to dimensions divisible by at least 8, if not 16, it will reduce scaling errors by the printer. Even better if you can provide image dimensions that are divisible by both 8 or 16 and at the printer's native ppi (not dpi - that's an entirely different subject) for the size you are printing.
 
Upvote 0
Apr 1, 2016
348
321
More than 33 MP. 8K DCI, which has been officially confirmed by Canon, is 8,192 pixels wide. Convert that to 3:2 and increase the width slightly so that both horizontal and vertical resolutions are divisible by 16 (for more efficient JPEG compression with fewer artifacts) and you get 8,208 x 5,472 = 44.9 MP. Plus the edge pixels needed for color interpolation on the edge of the output resolution and masked pixels.

Well, would be happy with that. 45 mpix sounds like a good allround resolution I could easily use for both events/weddings (maybe even in half sized mraw), and commercial and wildlife images for print. Still new to video so probably won’t touch 8k raw for now but 4k 10 bit log and full functioning dual pixel af even with 120 fps sounds great.
I hope they don’t forget the option to create 8k timelapse clips in-camera. Would be great for creating pans and zooms in post for final 4k output
 
Upvote 0

Michael Clark

Now we see through a glass, darkly...
Apr 5, 2016
4,722
2,655
Well, would be happy with that. 45 mpix sounds like a good allround resolution I could easily use for both events/weddings (maybe even in half sized mraw), and commercial and wildlife images for print. Still new to video so probably won’t touch 8k raw for now but 4k 10 bit log and full functioning dual pixel af even with 120 fps sounds great.
I hope they don’t forget the option to create 8k timelapse clips in-camera. Would be great for creating pans and zooms in post for final 4k output

None of the newer Canon cameras with .cr3 output that offer the C-Raw option have M-RAW or S-RAW options. Apparently M-RAW and S-RAW ended with .cr2.
 
  • Like
Reactions: 1 users
Upvote 0
It has nothing to do with compression rate. It has to do with compression efficiency

If you add padding pixels to align with 8x8 or 16x16 grid, you make the compression rate of unaligned images slighlty worse compared to the images that were aligned from the beginning.

They don't add any "fake pixels".

In jpeg compression, they do add padding pixels. Simply because jpeg algorithm is based on 8x8 chunks and not all the images are aligned to 8x8 chunks. Where does 16 come from I don't know, I trust you it's also related to jpeg, I'm not really into this matter tbh.

They add actual hardware photosites to the sensor outside the edges of the output resolution because color interpolation is done using surrounding photosites

Yes, but we were talking about the jpeg compression. You're now saying there are other technical reasons to align the sensor pixel dimensions to 16x16 - that's fine, my point was that
 
Upvote 0

Michael Clark

Now we see through a glass, darkly...
Apr 5, 2016
4,722
2,655
If you add padding pixels to align with 8x8 or 16x16 grid, you make the compression rate of unaligned images slighlty worse compared to the images that were aligned from the beginning.

If you have source files that are divisible by 8x8 or 16x16 you don't need to add any padding pixels, do you?

In jpeg compression, they do add padding pixels. Simply because jpeg algorithm is based on 8x8 chunks and not all the images are aligned to 8x8 chunks. Where does 16 come from I don't know, I trust you it's also related to jpeg, I'm not really into this matter tbh.

If you have source files that are divisible by 8x8 or 16x16 you don't need to add any padding pixels, do you?

Yes, but we were talking about the jpeg compression. You're now saying there are other technical reasons to align the sensor pixel dimensions to 16x16 - that's fine, my point was that

If you have source files that are divisible by 8x8 or 16x16 you don't need to add any padding pixels, do you?
 
Upvote 0

Stig Nygaard

EOS R7, Powershot G5 X II & Olympus TG-5
CR Pro
Jul 10, 2013
275
462
Copenhagen
www.flickr.com
They have comfirmed 8K RAW but not the specific resolution. There is no such thing as 8K DCI. The current DCI standards do not address 8K.

I think everyone agrees that if you are calling it "DCI", it means the horizontal resolution must be a multiply of 1024.
I'm surprised this is still discussed. It is as close to be confirmed as it can be, without actually having been officially confirmed. If it is a 3:2 sensor (and it would be a big surprise if it isn't), then it must be 44.7 megapixels.
 
  • Like
Reactions: 1 users
Upvote 0

Stig Nygaard

EOS R7, Powershot G5 X II & Olympus TG-5
CR Pro
Jul 10, 2013
275
462
Copenhagen
www.flickr.com
  • Like
Reactions: 2 users
Upvote 0

Michael Clark

Now we see through a glass, darkly...
Apr 5, 2016
4,722
2,655
I think everyone agrees that if you are calling it "DCI", it means the horizontal resolution must be a multiply of 1024.
I'm surprised this is still discussed. It is as close to be confirmed as it can be, without actually having been officially confirmed. If it is a 3:2 sensor (and it would be a big surprise if it isn't), then it must be 44.7 megapixels.

44.9 MP when you make it 8,208 x 5,472 so both sides of 3:2 still images are divisible by 16 like every other Canon EOS sensor ever made.
 
  • Like
Reactions: 1 users
Upvote 0

Stig Nygaard

EOS R7, Powershot G5 X II & Olympus TG-5
CR Pro
Jul 10, 2013
275
462
Copenhagen
www.flickr.com
44.9 MP when you make it 8,208 x 5,472 so both sides of 3:2 still images are divisible by 16 like every other Canon EOS sensor ever made.

Oh well, maybe. You're probably right :) I don't even know why we don't just write 45 megapixels. That's probably what Canon will call it anyway...
 
  • Like
Reactions: 2 users
Upvote 0
Apr 25, 2011
2,510
1,885
They don't add any "fake pixels". They add actual hardware photosites to the sensor outside the edges of the output resolution because color interpolation is done using surrounding photosites, both those that are filtered by the same color as the photosite that corresponds to the pixel on the edge of the output image as well as those that are filtered by different colors than the photosite in question. If a demosaicing algorithm uses three rows of photosites in all four directions to interpolate colors for the output pixel representing a particular photosite, then the sensor needs three rows of additional photosites on all four sides of the sensor beyond the number of rows and columns output in the final image. If a demosaicing algorithm uses five rows of photosites in all four directions, then the sensor needs five rows of additional photosites on all four sides, and so on.
Wouldn't those extra hardware pixels also need to appear in the RAW file data?
 
Upvote 0

koenkooi

CR Pro
Feb 25, 2015
3,575
4,110
The Netherlands
Wouldn't those extra hardware pixels also need to appear in the RAW file data?

RAW files already have 'additional' pixels, there's a border that is used for image housekeeping like black level calibration which doesn't show up in the resulting picture on your screen.

That's why Canon spec pages have:

Effective Pixels Approx. 30.3 megapixels
Total Pixels Approx. 31.7 megapixels

So that 44.9MP, is it the total or effective number of pixels?
 
  • Like
Reactions: 1 user
Upvote 0
If you have source files that are divisible by 8x8 or 16x16 you don't need to add any padding pixels, do you?

It's not such a big problem to add those padding pixels. Details of jpeg compression are unlikely to be the decisive factor for determining the sensor dimensions.

Some other factors could make them to be a multiple of 16, I don't argue that.

Demosaicing algorithms to do color interpolation have nothing to do with compression algorithms

But I never said they had.
 
Upvote 0
Uncompressed TIFF files are one form that represents these individual RGB values for each pixel.

btw, I don't think camera works with tiff files internally, the demosaicing process you mentioned is only needed for jpeg generation and back-screen previews, it's not written to raw files which leave demosaicing to the processing software.
Internally the camera would have buffers with RGB data per pixel, but it won't convert them to tiff files, it's kinda meaningless. tiff file implies more data than just RGB per pixel, and that data isn't necessary for interim processing.
 
  • Haha
Reactions: 1 user
Upvote 0
Mar 4, 2020
122
128
The "low light performance" like AF performance or the noise?
I'm extremely happy with both on my EOS-R. The only issue I have is the EVF when shooting very fast action, like BIF. For this reason I bought the Olympus EE-1 Dot Sight and that took care of the problem.
I was also comparing the 1-DX III to my first-generation 1-DX and while the Mark III files look better at higher ISO, the difference is not striking, if that's what you expect. I would say, maybe one stop difference.
The AF on the new 1-DX is vastly improved though, compared to the first generation.
Now, get this: yesterday I was playing with stacked teleconverters on my EOS-R and my 1-DX Mark III. The TC's are the latest Canon generation (1.4x and 2x).
Of course, I had to use a 12mm extension tube to stack them.

To my huge surprise, the EOS-R was way better to achive focus and to keep it. Go figure!
When, and that is a big "When", the 1-DX Mark III focused on the subject, it worked really well but the EOS-R was much, much more consistent and accurate.
If this is what the image sensor based AF means, then the R5 will be a beast of a camera.

Anyhow, the 1-DX Mark III AF is a pleasure to work with in any other conditions.
Did you try the 1DXiii in Live View mode? I believe that would allow it to use the same/similar focus system as the R5.
 
  • Like
Reactions: 1 user
Upvote 0
Now that we have a month's notice, maybe we should we start a forum discussion where we can start practicing excuses to justify the purchase to loved ones and concerned onlookers.

"Yes, I already have a camera, but I don't have that camera!"

Totally unrelated but I`m wondering what´s the fuss with buying the camera you want with your own money. I keep reading all this comments of asking "permission" to your wife to use your own money. Seems so ridiculous to me.

Man up people.

Back to photography, I really hope the R6 has sensor IS and is priced around 2000$, that will be great and I will buy it to replace my old 6D and my 80D.
 
Upvote 0
Mar 4, 2020
122
128
Or it was argued that 'DCI' only applies to '4K' part of '8K/4K DCI'. So it reads as '8K and 4K DCI', not '8K DCI and 4K DCI'.

Because of that, I'm still not 100% convinced the camera will have 45Mp. I really hope it will, I'll probably make a preorder in this case. 39Mp? meh, I don't know, will probably wait for R5s.
But if they don't make the sensor for 8k DCI, wouldn't that make the down-sampling to full width 4K DCI much more difficult? If the sensor is built at 8K DCI resolution (i.e. double 4K in width and height) I would think the algorithm would be much easier and less processor intensive...
 
  • Like
Reactions: 1 users
Upvote 0
The "low light performance" like AF performance or the noise?
I'm extremely happy with both on my EOS-R. The only issue I have is the EVF when shooting very fast action, like BIF. For this reason I bought the Olympus EE-1 Dot Sight and that took care of the problem.
I was also comparing the 1-DX III to my first-generation 1-DX and while the Mark III files look better at higher ISO, the difference is not striking, if that's what you expect. I would say, maybe one stop difference.
The AF on the new 1-DX is vastly improved though, compared to the first generation.
Now, get this: yesterday I was playing with stacked teleconverters on my EOS-R and my 1-DX Mark III. The TC's are the latest Canon generation (1.4x and 2x).
Of course, I had to use a 12mm extension tube to stack them.

To my huge surprise, the EOS-R was way better to achive focus and to keep it. Go figure!
When, and that is a big "When", the 1-DX Mark III focused on the subject, it worked really well but the EOS-R was much, much more consistent and accurate.
If this is what the image sensor based AF means, then the R5 will be a beast of a camera.

Anyhow, the 1-DX Mark III AF is a pleasure to work with in any other conditions.

Repeat te test in LiveView for the 1dx and you will be surprised.
 
Upvote 0