Canon EOS R5 Mark II firmware update coming this month

#41
David - Sydney said:
Clearly we aren't entering prompts but there are localised image generation models
Click to expand...
That's not LLM. LLM stands for "large language model" and it has a specific meaning.

As I said, "AI" in this case most likely means some neural network processing, obviously trained on images - but noise reduction or upscaling is not image generation, it's fine detail generation so to say.
 
  • Like
Reactions: 1 user
Upvote 0
#42
  • Haha
Reactions: 3 users
Upvote 0
#43
Quarkcharmed said:
Literally all AI developed by humanity so far is algorithmic. Even the most powerful ChatGPT versions are just programs with a lot of data. The borderline between what is "AI" and what is not "AI" is a little bit fuzzy.

With camera manufacturers, "AI" most likely means the use of artificial neural networks in some form.
Click to expand...
The entire term A.I. is pretty fuzzy.
I think people are under the impression that A.I. should be able to learn on its own and as I type this, no such A.I. exists.
Machine learning is nothing new and it is used all over the place.
If anything, things that have been quietly using ML all along have added A.I. to their names to capitalize on the hype of ChatGPT.
That is not to say that some are not lying about it.
Many are not.
 
  • Like
Reactions: 1 user
Upvote 0
#46
EOS 4 Life said:
The entire term A.I. is pretty fuzzy.
I think people are under the impression that A.I. should be able to learn on its own and as I type this, no such A.I. exists.
Machine learning is nothing new and it is used all over the place.
If anything, things that have been quietly using ML all along have added A.I. to their names to capitalize on the hype of ChatGPT.
That is not to say that some are not lying about it.
Many are not.
Click to expand...
DeepMind’s later generation chess, go etc were done purely on being fed with the rules of the games and then playing against itself with self-reinforcement. The programs learned by themselves
 
Upvote 0
#48
AlanF said:
DeepMind’s later generation chess, go etc were done purely on being fed with the rules of the games and then playing against itself with self-reinforcement. The programs learned by themselves
Click to expand...
That's impressive, considering that Deep Thought took 7.5 million years to determine that the Ultimate Answer was 42.
 
  • Like
  • Haha
Reactions: 1 users
Upvote 0
#49
peters said:
woah, what? 4 times it froze in such a short period of usage? Woah, that's intense!!! In a demanding environment, or during event documention with your client in the neck that's certainly a realy realy bad thing to happen. This would be way to much for me... that sounds like a heavy problem to me.

Both my r5 only froze in the very beginning due to overheating (which is bad enough, but once I learned in which modes it happened, I worked around it), but never at random...
Click to expand...
Yeahhh, it’s not great, that’s for sure. But the good news is that it has only happened between recording and not during recording. I was shooting 4K60 and maybe about 45 minutes into shooting clips, I got this screen. I couldn’t get it to happen again that day. But, it would happen 3 more times over the course of my use.

New BUG today: shot 3 photos in a row at 5fps in electronic shutter and the screen went black (made the electronic shutter sound for each shot) those 3 photos never recorded and they weren’t on the memory card.

Keep in mind, I’ve had this camera since it shipped last month. I’ve shot 5 motorsports events with it as an 3rd camera or used in the pits and have roughly 15,000 photos on one of the bodies already. I’m actually surprised I haven’t discovered more problems. My original R5 froze OFTEN until a few firmware updates and has been flawless ever since. My R3 to this day will “once in a blue moon” freeze or have an error. I use those bodies almost every single day to shoot photo and video in a pretty demanding environment and consider them totally reliable.

The R5II hasn’t thrown an error code yet. Just a few freezes and small quirks that have come up…like being EXTREMELY slow to turn on when a 1TB card from ProGrade or Angelbird is inserted, but will instantly turn on when a 325GB ProGrade is inserted.

These are the growing pains of being an early adopter. I never show up anywhere with fewer than 3 camera bodies and lots of memory cards and batteries. I’m mixing the R5II in where I can and it has been working well. I wouldn’t be too concerned. It’s going to be an absolute powerhouse for video production and photography - no doubt.
 

Attachments

  • IMG_6359.jpeg
    IMG_6359.jpeg
    2.7 MB · Views: 15
  • Like
Reactions: 1 user
Upvote 0
#50
neuroanatomist said:
That's impressive, considering that Deep Thought took 7.5 million years to determine that the Ultimate Answer was 42.
Click to expand...
The DeepMind chess program came up with new ideas in chess from playing millions of games against itself. I was invited a couple of years ago to write a paper on AlphaFold, the revolutionary program from the same company that predicted the structure of proteins. I couldn't help but thrown in some comparisons with chess programs and camera AF. ;)
 
Upvote 0
#52
rbielefeld said:
You can do this. I use a small flexible zone with Eye Detection set to Auto and it will lock on the eye and only track the eye within the zone. Also, you can set the AF area to Spot AF and have Eye Detection set to Auto and it will detect the eye of a small bird at the AF Spot location and not track it throughout the entire frame.
Click to expand...
Thanks! This helped me find a solution that works for me with a R5m2.
For small birds in trees I find it essential to be able to toggle back and forth between i) focus on/near the bird with spot focus and ii) eye focus on/near the bird with spot focus. I figured out one way to do it by setting
Pink Menu
1 Whole area tracking = OFF
2 Eye detection = Auto
2 Subject to detect = Animals
5 Limit AF areas to spot and spot-locked
5 Limit subject to detect bird
Then I assigned buttons to enable i) half press of the shutter button for metering and focus, ii) AF-ON button to toggle eye detect on/off, and iii) full press of the shutter button to take pictures.

The button setup is a bit awkward in that it is almost the inverse of my usual back-button focus, but it does enable me to toggle eye focus on and off during servo AF with my forefinger and thumb remaining on only 2 buttons. Has anyone found a better way to do this?
 
Last edited:
Upvote 0
#53
I'd like a auto white balance lock option to ensure entire video clip has the same balance. Also the fan grip fan speed menu needs a temperature feedback loop option, like a thermostat found on just about every other modern device with a cooling fan.
 
Upvote 0
#55
Quarkcharmed said:
That's not LLM. LLM stands for "large language model" and it has a specific meaning.

As I said, "AI" in this case most likely means some neural network processing, obviously trained on images - but noise reduction or upscaling is not image generation, it's fine detail generation so to say.
Click to expand...
Yep, it won't be a LLM locally but local GenAI is coming/already here. How it is trained is the interesting part. Apple Intelligence (oxymoron?) seems to be an interesting application where requests aren't sent the cloud for generation but trained on local images. The choice in the future to train a local genAI using our Lightroom catalogues will be a good discussion.

It is probably unlikely that Canon was able to incorporate that level of "generative" intelligence in the R5ii as it is pretty bleeding edge although the co-processor seems to be designed for that purpose (and AF processing). Maybe a firmware update in the future??

It would be a great white paper to read exactly how Canon has implemented the upscaling algorithm as we can only guess and probably ignore the feature otherwise.
 
Last edited:
  • Like
Reactions: 1 user
Upvote 0
#56
David - Sydney said:
Yep, it won't be a LLM locally but local GenAI is coming/already here. How it is trained is the interesting part. Apple Intelligence (oxymoron?) seems to be an interesting application where requests aren't sent the cloud for generation but trained on local images. The choice in the future to train a local genAI for our Lightroom catalogues will be a good discussion.

It is probably unlikely that Canon was able to incorporate that level of "generative" intelligence in the R5ii as it is pretty bleeding edge although the co-processor seems to be designed for that purpose (and AF processing). Maybe a firmware update in the future??

It would be a great white paper to read exactly how Canon has implemented the upscaling algorithm as we can only guess and probably ignore the feature otherwise.
Click to expand...
Canon global has a webpage describing Deep Learning image technology: https://global.canon/en/technology/dl-iptechnology-2023.html

The links to two whitepapers are at the bottom of the page.
 
  • Like
Reactions: 4 users
Upvote 0
#57
P-visie said:
Canon global has a webpage describing Deep Learning image technology: https://global.canon/en/technology/dl-iptechnology-2023.html

The links to two whitepapers are at the bottom of the page.
Click to expand...
Thanks! Not genAI in theory as ML only. I don't know what "made with AI" label would be affixed or not.
I do wonder if it would be better in DPP rather than in-camera though as it seems to be a manual and time consuming process although it does get added to the body spec sheet.

I wonder if Canon's "immense image database" came from body beta testers in the field as part of the testing.
Copyright stays with the tester of course but Canon gets a copy for its internal usage.
I'm not sure that Canon would have access to "every sort of subject possible" otherwise.

The R5ii's scaling doubles vertical/horizontal pixel count but only works on jpg or HEIF images not raw. It seems strange that anyone would choose from 8 bit jpg. It seems that jpgXL format is not supported unfortunately.

For other deep learning corrections, I'm surprised that the astro/coma improvement fixed the blown highlight but doesn't really fix the coma aberration. The Moire one looks useful though!
 
Upvote 0
#58
David - Sydney said:
Thanks! Not genAI in theory as ML only. I don't know what "made with AI" label would be affixed or not.
I do wonder if it would be better in DPP rather than in-camera though as it seems to be a manual and time consuming process although it does get added to the body spec sheet.

The R5ii's scaling doubles vertical/horizontal pixel count but only works on jpg or HEIF images not raw. It seems strange that anyone would choose from 8 bit jpg. It seems that jpgXL format is not supported unfortunately.
Click to expand...
Machine learning is a subset of AI, so, as others have stated, the ‘AI’ label can be attached like the ‘Turbo’ of earlier times.

Apart from taking to much processor capacity and time to upscale RAW in camera, I suspect that this feature was developed for the R1, where (some) Pro’s have a jpg workflow for working under tight timescales, and the features is inherited by the R5 Mk II.

Like IBIS picture shift, upscaling is, except for specific uses cases, a ‘gimmick’ for most R5 (Mk II) users.
 
  • Like
Reactions: 2 users
Upvote 0
#59
P-visie said:
Canon global has a webpage describing Deep Learning image technology: https://global.canon/en/technology/dl-iptechnology-2023.html

The links to two whitepapers are at the bottom of the page.
Click to expand...
Interestingly, they compare it to interpolation instead of competing out-of-camera software.
In-camera makes sense because of the convenience but the out-of-camera version requires a paid subscription.
The competition is not free either but we need to see how they stack up.
 
  • Like
Reactions: 1 users
Upvote 0
#60
EOS 4 Life said:
Interestingly, they compare it to interpolation instead of competing out-of-camera software.
In-camera makes sense because of the convenience but the out-of-camera version requires a paid subscription.
The competition is not free either but we need to see how they stack up.
Click to expand...
I agree. Out-of-camera requires the use of DPP, that will be an (in addition to the paid subscription) obstacle for widespread use.
 
Upvote 0
You must log in or register to reply here.

Similar threads

Canon Rumors
The next Canon EOS R5 Mark II firmware update coming this month
Replies
56
Views
6K
alFR
A
Canon Rumors
Canon has released Canon EOS R5 Mark II Firmware v1.0.2
Replies
46
Views
3K
Del Paso
D
Canon Rumors
Update – Canon EOS R5 Mark II Specifications
Replies
118
Views
21K
neuroanatomist
neuroanatomist
Canon Rumors
Canon EOS R5 Mark II firmware v1.0.1 released
Replies
76
Views
8K
AlanF
AlanF
Canon Rumors
Canon EOS R1 & Canon EOS R5 Mark II announcement coming July 17
Replies
342
Views
16K
Michael Clark
Michael Clark
Top Bottom
Back
Home
Forums
Media
Menu