Canon EOS R7 pre-sales discussion, hypothesizing, and prognosticating

John Sheehy wrote in post #19390788
I haven't looked at the full set of ISOs, but I downloaded two ISO 1600 CR3 samples from DPR (90D and R7) and looked at the masked pixels on the left in RawDigger, and the R7 does have a lower standard deviation of noise by 23%, but that could be for other reasons than less noise, like RAW cooking, or assigning more highlight headroom to the RAW levels. More work is needed to see why it is lower, but if it is really lower noise and nothing else, then the R7 may have one of the best sensors out there for noise per unit of sensor area.

I personally believe Canon has started to cook the raw just a bit in recent models to help the results along, but I can't prove it. It just seems raw results from newer bodies already seem to be a bit "processed" to me. I guess it doesn't matter as long the final results still has all the detail needed for the final result.

10 days to go.....let the countdown begin .

According to Bill Claff at Photons-to-photos, both the R5 and R6 have been 'cooked' at lower ISOs, and the R3 throughout the ISO range.

As long as there is no pushback from users, further cooking is likely to occur. Can't say that is good or bad as far as most users are concerned, but the RAW files aren't really what I would call 'raw' anymore.

Will be interesting to see if other manufacturers follow.

Peter2516 wrote in post #19390874
10 days to go.....let the countdown begin .

Fingers crossed for all of the day one preorders!

Hopefully, they’ll post the user manual soon too.

other manufacturers do it a long time ago

When they fabricate noise reduction circuitry onto the sensor, everybody says it’s a good feature. But when they do the exact same algorithm in the camera firmware, people get indignant and accuse them of “cooking” the raw data.

Fundamentally, is there any difference?

for people using Animal AF here is another example from my 90D, this was not on 1 point but on 9 point, trees were at 100 yards, duck was probably about at 80 yards so it is a small target against a bad background. How do the R5 R6 and R3 handle this, especially the R3?
Am I hoping for too much? but this is my usual background for the most part and as I said I am going to rent and try it real life.

Image hosted by forum (1163623) © Jeff USN Photog 72-76 THIS IS A LOW QUALITY PREVIEW. Please log in to see the good quality stuff.

mcoren wrote in post #19391041
When they fabricate noise reduction circuitry onto the sensor, everybody says it’s a good feature. But when they do the exact same algorithm in the camera firmware, people get indignant and accuse them of “cooking” the raw data.

Fundamentally, is there any difference?

There's a huge difference between detecting and cancelling camera-generated readout noise as it is actually happening, vs looking at an already-captured image and filtering it; the latter loses fine signal, the former doesn't.

Jeff USN Photog 72-76 wrote in post #19391048
for people using Animal AF here is another example from my 90D, this was not on 1 point but on 9 point, trees were at 100 yards, duck was probably about at 80 yards so it is a small target against a bad background. How do the R5 R6 and R3 handle this, especially the R3?
Am I hoping for too much? but this is my usual background for the most part and as I said I am going to rent and try it real life.

Hosted photo: posted by Jeff USN Photog 72-76 in
./showthread.php?p=193​91048&i=i79276000
forum: Canon Digital Cameras

Yes, you have high hopes. I don't think the R7 is going to just grab the bird 100% of the time; you need to help, perhaps. If you can pan the bird perfectly that will help, as the bird will then have more contrast than the blurred background, so any time you "stop" panning, AF is more likely to grab the trees. If the AF is already on the trees, and the bird is far enough out of focus, then it could get stuck on the background, but as per your example, the bird may have been close enough to focus for the AI to identify it, and stick to it. Probably a borderline case there, with your example, that could work or could require some help, by pre-focusing on the close side of the bird. IOW, you could pick a tree off to the side that is closer than most of the birds, and AF on that first, when waiting for new BIFs to appear. Or, if your focus button setup allows, or the lens has full time manual focus override, then you could learn to turn the focus ring to keep getting off of the background. This is based on R5 experience; the R7 may be better in this regard and see subjects better through initial moderate defocus.

mcoren wrote in post #19391041
When they fabricate noise reduction circuitry onto the sensor, everybody says it’s a good feature. But when they do the exact same algorithm in the camera firmware, people get indignant and accuse them of “cooking” the raw data.

Fundamentally, is there any difference?

There is a difference in process and technique in determining noise at the hardware level (especially for noise introduced by the electronics themselves) vs software to try to figure out noise from the image detail before writing "raw" data to a file.

Software algorithms for noise removal evolves annually, and that kind of noise reduction belongs outside the camera IMO. Sometimes I might be able to remove noise better than some canned software function that was added to the firmware at the time the code was assembled for camera productions.

As we really don't know what and where this is happening in all reality, we just know we are getting an altered "raw" file, all we can discuss is "whether the end user is more capable here than something in-camera". I say yes for me personally, but probably not the answer for the general population.

Also what this means is that there are fewer and fewer gains to hardware changes, and to show "evolution", Canon has to do more with software in the firmware.

Jeff USN Photog 72-76 wrote in post #19391048
for people using Animal AF here is another example from my 90D, this was not on 1 point but on 9 point, trees were at 100 yards, duck was probably about at 80 yards so it is a small target against a bad background. How do the R5 R6 and R3 handle this, especially the R3?
Am I hoping for too much? but this is my usual background for the most part and as I said I am going to rent and try it real life.

Hosted photo: posted by Jeff USN Photog 72-76 in
./showthread.php?p=193​91048&i=i79276000
forum: Canon Digital Cameras

The R5/R6 recognize objects and tracks those objects. The phase detect AF board on the 90D cannot do this, for multiple reasons. However, in Liveview, you should have face/object tracking (called iTR) that might be able to follow that bird for you better than you trying to in the viewfinder. Different AF system entirely in liveview... (sensor is the AF system, not the separate PDAF AF board elsewhere in the camera).

You are fighting multiple issues here.
1) Distance (and thus size) of the bird you want to track. The larger you can get that bird into the frame either with a longer lens or use of TCs, the better.

2) The dark and busy background... that is going to cause some shifting of AF to that from time to time if you cannot track the bird manually with a set AF point, or if the camera is set up to try to locate objects for you.

3) PDAF system is vastly different on the DSLRs than the sensor based PDAF or contract detect AF. Mirrorless expands on the the sensor based AF model, with better AF algorithms and speeds than with DSLRs, other than the 1DX3 which has basically the same "mirrorless" solution in Liveview as the R6.

4) The smaller the object, the more you are digitally zooming into the bird after the fact, and the right camera with the right lens to be able to do that to still keep good detail and a decent version of the bird will be quite expensive. This is due to need of resolution, lower AA filter strength, and a very good high resolving lens.

5) With those kinds of distances, almost every single time you are fighting atmospheric distortion, and there isn't anything you can do to combat that other to get closer (reduce the space for convection currents to ruin the scene). So even if the R7 is the absolute best camera ever, and you put a 500mm or 600mm big white on that camera, your results will still be soft due to this, so you cannot throw money at this factor, short of buying a boat and getting closer to all the wildlife.

Points 1, 4 and 5 are all related to each other from different perspectives.

John Sheehy wrote in post #19391054
There's a huge difference between detecting and cancelling camera-generated readout noise as it is actually happening, vs looking at an already-captured image and filtering it; the latter loses fine signal, the former doesn't.

TeamSpeed wrote in post #19391064
There is a difference in process and technique in determining noise at the hardware level (especially for noise introduced by the electronics themselves) vs software to try to figure out noise from the image detail before writing "raw" data to a file.

Software algorithms for noise removal evolves annually, and that kind of noise reduction belongs outside the camera IMO. Sometimes I might be able to remove noise better than some canned software function that was added to the firmware at the time the code was assembled for camera productions.

As we really don't know what and where this is happening in all reality, we just know we are getting an altered "raw" file, all we can discuss is "whether the end user is more capable here than something in-camera". I say yes for me personally, but probably not the answer for the general population.

Also what this means is that there are fewer and fewer gains to hardware changes, and to show "evolution", Canon has to do more with software in the firmware.

I’m nowhere near as knowledgeable as most folks here in noise and mitigation, but I am familiar with real-time embedded processor systems.

Don’t assume that the in-camera algorithm is the same one that Lightroom or Photoshop uses. If that was all they did, I agree that would be weak.

But the camera firmware has access to many real-time sensor properties that the PP software does not, because they’re not included in the raw file. For example, there are many “dark” photosites on the sensor that can be used to estimate the noise in real time, based on the sensor temperature and its specific fabrication tolerances. Such an algorithm could potentially be much more accurate than the “one size fits all” sliders in Photoshop.

None of us really know what the firmware is doing.

I can guarantee that if you take the best canon camera and apply the best settings you can to get the best jpg through the canon firmware, I can take the raw and do better. I have gone through generation after generation of canon to develop NR regiments, each time better than anything canon has in body.

I am not the only one, many here have seen the same thing, so I am confident saying we collectively do know quite a bit about canon NR functionality.

TeamSpeed wrote in post #19391086
I can guarantee that if you take the best canon camera and apply the best settings you can to get the best jpg through the canon firmware, I can take the raw and do better. I have gone through generation after generation of canon to develop NR regiments, each time better than anything canon has in body.

I am not the only one, many here have seen the same thing, so I am confident saying we collectively do know quite a bit about canon NR functionality.

No argument here. I know I’m waaaaayy out of my league discussing noise with you or John or any of a thousand regular contributors here.

My point is that Canon knows the people who shoot raw already know how to do that in post. if they’re doing that processing in the camera firmware, I’ve got to believe it’s because the firmware in the camera can make corrections based on information that’s no longer available once the raw file is written.

Otherwise, why stop there? Why not do sharpening and color and temperature and histogram correction too? Before long, the raw file is nothing more than a high quality JPEG.

Found these on the net

25000 ISO files on the R7.....Not usable for serious results in my opinion

here is the link

Im going to hold off placing an order until we can see raws. 6400 might be tough

https://www.photograph​yblog.com/previews/can​on_eos_r7_photos