Here are more images of the Canon EOS R3

[freejay]

I have this idea that the R3 will be able to track a specific subject, e.g. a dedicated person ('s face) like the bride on a wedding. This (from Canon USA) sounds a bit like it (last sentence):

The EOS R5 and R6 were only the beginning.
Using Deep Learning technology,
the upcoming EOS R3 will offer enhanced AF performance
and tracking capabilities, with even better face-, eye-,
head- and body-detection.
And, the EOS R3 will add a new type of subject recognition, for its AF system.

 

[neuroanatomist]

I love flippy screens but unfortunately I love L brackets. It is a tough life I live.

The two are not incompatible. I have an RRS L-bracket on my EOS R. It constrains the movement a bit, but not excessively.

I do hope the RRS bracket for the R3 is modular like the one for my 1DX, the bracket for the EOS R is a single piece.

 

[SereneSpeed]

I still don't understand why Canon did not bother at least to "weld" a control ring adapter instead of just a basic one. These are $12K lenses and even the ultra cheap 50mm 1.8 has a control ring.

Ergonomics? Those are long lenses. I can’t see wanting my left hand to come all the way back towards the body to adjust the control ring, while using lenses that long. The new bodies have enough controls to adjust the exposure triangle with your right hand. Even with the much shorter 100-400ii, it feels uncomfortable to adjust the control ring. For me at least.

 

[Photo Bunny]

I was hoping for a fixed screen or a flip out. Articulating screens are a bunch of extra steps to get a low shot.

 

[privatebydesign]

I have this idea that the R3 will be able to track a specific subject, e.g. a dedicated person ('s face) like the bride on a wedding. This (from Canon USA) sounds a bit like it (last sentence):

The EOS R5 and R6 were only the beginning.
Using Deep Learning technology,
the upcoming EOS R3 will offer enhanced AF performance
and tracking capabilities, with even better face-, eye-,
head- and body-detection.
And, the EOS R3 will add a new type of subject recognition, for its AF system.

I think we are years from that kind of scenario in retail cameras, maybe high end security cameras but not regular cameras.

I think you are reading too much into Canon's statements, the R5 and R6 do not 'learn', they were taught/programmed with a large number of possible scenarios, they never learn more nor refine their processes as they get more experience. That is not 'learning' and it is not Ai, it is marketing bull...

 

[Adelino]

I think we are years from that kind of scenario in retail cameras, maybe high end security cameras but not regular cameras.

I think you are reading too much into Canon's statements, the R5 and R6 do not 'learn', they were taught/programmed with a large number of possible scenarios, they never learn more nor refine their processes as they get more experience. That is not 'learning' and it is not Ai, it is marketing bull...

They can be "taught" with updates though? Is that something that happens?

 

[dcm]

Ergonomics? Those are long lenses. I can’t see wanting my left hand to come all the way back towards the body to adjust the control ring, while using lenses that long. The new bodies have enough controls to adjust the exposure triangle with your right hand. Even with the much shorter 100-400ii, it feels uncomfortable to adjust the control ring. For me at least.

The control ring on the RF800 is just beyond the focus ring where you would want it, not next to the body. This might require major modifications to the existing EF lenses.



I attached 2" ARCA plate to the tripod mount, extending forward under the focus ring so the lens weight is not on the rings when handheld. The plate rests in the palm of my hand, with easy access to both the focus ring (ring finger, pinkie) and control ring (index finger, middle finger). I do find the control ring quite helpful for exposure compensation when shooting birds. With eye-AF, I don't seem to use the focus ring that often.

 

[gregster]

I think we are years from that kind of scenario in retail cameras, maybe high end security cameras but not regular cameras.

I think you are reading too much into Canon's statements, the R5 and R6 do not 'learn', they were taught/programmed with a large number of possible scenarios, they never learn more nor refine their processes as they get more experience. That is not 'learning' and it is not Ai, it is marketing bull...

I don't think it's that far fetched. The way it's described sounds similar to Sony's real-time tracking mode, where objects are tracked without requiring eye/head detection.

 

[SteveC]

They can be "taught" with updates though? Is that something that happens?

Even if it did, it'd be a far cry from a camera that learned from its own experience, which would be useful in that it could eventually figure out how you yourself want to handle a situation (though it would be maddening if you are more unpredictable or are yourself learning, because it would think the thing you did six months ago is what you want to do now and you'd constantly have to undo that).

And it can only be 'taught' what Canon is aware it needs to know.

 

[privatebydesign]

They can be "taught" with updates though? Is that something that happens?

Well it depends on your definition of taught. If you mean it can have more random scenarios programmed in then yes, but I don't think that is most peoples idea of learning. That would be closer to gaining unique insight due to specific individual use and changing the actions based on the results of those new experiences. How could a camera do that if the results of its actions are never graded, it doesn't 'know' when it did better so it can't 'learn'.

 

[privatebydesign]

I don't think it's that far fetched. The way it's described sounds similar to Sony's real-time tracking mode, where objects are tracked without requiring eye/head detection.

That is 'an object' not a self determined specific subject. What was proposed is a scenario where, for instance, a bride and father were walking down an aisle and the camera 'knew' the bride from the father. Nothing in the regular camera scene can do that yet, so far the camera knows there are two subjects walking towards it but it prioritizes those two subjects due to pre determined and dumb criteria. Normally the closest subject gets priority.

I think we are years away from a point where regular cameras can face detect, that is, you could take a scan of the bride while getting dressed and then for the rest of the day the camera will scan any face in the frame and prioritize AF to her. I can see it happening it time, but we are years away from it being commercially available or affordable.

When it does happen it will happen in phones first because they can already recognize individual faces.

 

[EOS 4 Life]

it could be that the articulated lcd screen makes the R3 a notch 'less pro' than the R1

I can imagine a pro-level articulating screen but the hinge would need to be more rugged.
A fixed screen seems almost pointless on a DSLM since photographers would mostly use the EVF and videographers prefer flip screens.

 

[EOS 4 Life]

Your phone screen is glued to the frame and doesn't flip out. A bezel helps maintain rigidity and also protects the screen edges.

Maybe, but none of that excuses the enormous size of the camera bump when compared to my phone.

 

[EOS 4 Life]

You clearly aren't capable of absorbing the information that the R3 screen hinge could have been omitted and the same space used for a larger fixed screen.

These are to scale. The distance from the edge of the R3 to the actual screen edge is much larger than the distance from the edge of the 1DX II to the viewable screen edge.


View attachment 197900 View attachment 197901

You are kind of both right but pointlessly arguing.
The reason that the R3 screen could be bigger is that it needs its own bezels for durability.
If it had no bezels then the space taken up the hinge would be negligible.
An R1 would be expected to be even more durable so it would require an even bigger hinge and even bigger bezels.
On the other hand, a flip-screen adds extra durability when flipped closed.

 

[dcm]

They can be "taught" with updates though? Is that something that happens?

This article is a nice introduction to Artificial Intelligence and Machine Learning in cameras.

TL;DR
The jist of the article is that the learning is done in the lab using a collection of sample images to determine the configuration of the neural network. The finished neural newtork is loaded into cameras. The camera doesn't have the processing power to learn, but it can use a network that has already been learned.

Each sensor and digic combination will likely require a different neural network. Sensors will have different characteristics and digics will have different processing capabilities. We see this in newer processors like the Apple M1 which includes a neural engine on the chip. Incremental mprovements to the neural network could come from training with additional images and downloaded via firmware. Large improvements will come from new system/digic combinations.

If you want more detail, there is a recent paper in the IEEE Transactions on Computational Imaging entitled Deep Learning for Camera Autofocus.

 

[EOS 4 Life]

Don’t think canon will have a mechanical shutter on a camera that is global shutter. It seems redundant.

A mechanical shutter consumes less energy and protects the sensor from dust.
On the other hand, thanks to Kolari we know there is room for internal ND and IBIS together in the RF mount.

Kolari R Magnetic Clip-In Filter for Canon RF Mount

We’re excited to introduce our new R mount clip filters as well! Compatible with the R adapter, or native RF lenses, these can be used to add rear filtering to RF lenses, or two layers of filtering when paired with the R filter adapter. The clip filter can be used to quickly change the effective...

kolarivision.com

It is possible that removing the mechanical shutter would leave room for stacking two.
The best would be mechanical shutter + internal ND + IBIS.

 

[privatebydesign]

You are kind of both right but pointlessly arguing.
The reason that the R3 screen could be bigger is that it needs its own bezels for durability.
If it had no bezels then the space taken up the hinge would be negligible.
An R1 would be expected to be even more durable so it would require an even bigger hinge and even bigger bezels.
On the other hand, a flip-screen adds extra durability when flipped closed.

Er, no. Either hinges exist in space or they don't. If they do exist then that means something else can't exist in that space, if they don't exist then it means something else can occupy that space. There is no "both right" to that.

My two images perfectly illustrate the concept and practical use of... hinges.

 

[EOS 4 Life]

They can be "taught" with updates though? Is that something that happens?

That only works if there is some sort of feedback like Tesla has with Autopilot.
The car itself does not learn but it records its mistakes.
Tesla compiles all the mistakes from each car.
Then when common solutions are found they push them out to all of the cars.
The only feedback Canon has is when we report errors.

 

[dcm]

That is 'an object' not a self determined specific subject. What was proposed is a scenario where, for instance, a bride and father were walking down an aisle and the camera 'knew' the bride from the father. Nothing in the regular camera scene can do that yet, so far the camera knows there are two subjects walking towards it but it prioritizes those two subjects due to pre determined and dumb criteria. Normally the closest subject gets priority.

I think we are years away from a point where regular cameras can face detect, that is, you could take a scan of the bride while getting dressed and then for the rest of the day the camera will scan any face in the frame and prioritize AF to her. I can see it happening it time, but we are years away from it being commercially available or affordable.

When it does happen it will happen in phones first because they can already recognize individual faces.

Object identification has improved a lot as we see with the people/animal and eye tracking options. The camera may already detect mutiple objects like the smartphones, but only shows you the one selected for AF based on its current rules. What's missing is a method to let you select which object(s) you would like to maintain in focus when multiple objects exist. A joystick approach may work for this. Paired with eye-controlled focus on the M3 to select an object, this might be kind of interesting. Instead of moving from focus zone to focus zone, the AF would simply move object to object - the objects become the AF zones. And this might make the R3 much more interesting.

 

[maulanawale]

I have always been surprised that Olympus does not sell many more cameras than what they do. They make some great products and have a cult like following.

To be honest, I think their main problem has always been their disastrous marketing strategy. Granted the system has shortcomings, but it still perfectly capable and in some use cases, more convenient than FF set ups. Having both is the ultimate kit for me.

I have an Olympus. I love the way it looks. For me, that's where the attraction ends. Menus? OMG!!!!

lol yep, but after using Sony, even Olympus is a breath of fresh air in terms of menus. I’m very partial to the EM1X anyway mainly because I enjoy using it like a kid with a new toy.
Sorry for the off topic, I came here to geek out on the R3