PhotonsToPhotos does the Canon EOS R5 Mark II and it’s good

[vikingar]

At least according to the Nature news article that I linked above, the highest rate of non-reproducibility is among chemists, followed by biologists then physicists/engineers.

View attachment 219070

That Nature news article has to be off, or at least heavily biased towards Nature's fields. Isn't it famously psychology that is almost entirely irreproducible? See e.g. here: https://en.wikipedia.org/wiki/Replication_crisis

I would guess math is probably the most reproducible. However I know from personal experience that even a strongly related field like computer science already has tons of papers with factual errors, non-reproducable or outright wrong results.

 

[AlanF]

That Nature news article has to be off, or at least heavily biased towards Nature's fields. Isn't it famously psychology that is almost entirely irreproducible? See e.g. here: https://en.wikipedia.org/wiki/Replication_crisis

I would guess math is probably the most reproducible. However I know from personal experience that even a strongly related field like computer science already has tons of papers with factual errors, non-reproducable or outright wrong results.

Our 2 posts came at about the same time. It was important, as always, to look at the data behind the study and not just the claimed results.

 

[HMC11]

Thanks to AlanF & vikingar for pointing this out. It's particularly galling for Nature to carry this given its reputation for quality, which can lead readers to give the published works a pass at face value. I guess Nature News is a different matter altogather, but it does sully the good name of Nature.

 

[vignes]

I have the A1 and getting the R5II. I\'m sorry CR, the Sony eco system especially the lens option is ahead of RF system. The A1 is expensive because Sony released a similar camera early with a higher readout speed than Canon. You pay to be an early adaptor. Some of Sony\'s pro lens (GM) and 3rd party are very good plus they\'re small/compact with respect to the performance. Canon at this stage can\'t match that. There is another consideration, all Sony lens can be adapted to Z mount cameras. I have been testing my GM lenses on Z8 using the Megadap adaptor. Excellent performance. I know with RF lens investment, I can\'t do this.
I do have some RF and EF lens and thinking of getting the R5M2. There is no rental option now and I may have to buy one. I was one of the earliest to buy the R5 and I had to return the camera one week after purchasing it. It kept shutting down on overheat issue and the retailer couldn\'t solve it. I got a full refund. I know after couple of months later, Canon managed to reduce the issue via FW updates. I think the R5M2 would be a better designed camera as Canon would have learned the lesson. I want to give another try.

 

[AlanF]

Thanks to AlanF & vikingar for pointing this out. It's particularly galling for Nature to carry this given its reputation for quality, which can lead readers to give the published works a pass at face value. I guess Nature News is a different matter altogather, but it does sully the good name of Nature.

Scientific fraud is a serious problem. Harder data of where it occurs may be found in Retraction Watch which lists retracted papers. They have a “Leader Board” of the most retracted authors. Anaesthesiology is the clear winner. https://retractionwatch.com/the-retraction-watch-leaderboard/

 

[Raindropdroptop]

Here you are: The R6ii looks significantly lower at lowest iso in ES and no difference in mechanical, with the caveats in my last post.

View attachment 219055

Looks like Canon cranked up the built in raw noise reduction.

 

[AlanF]

Thanks to AlanF & vikingar for pointing this out. It's particularly galling for Nature to carry this given its reputation for quality, which can lead readers to give the published works a pass at face value. I guess Nature News is a different matter altogather, but it does sully the good name of Nature.

Something else to think about. Nature itself has a very high proportion of retracted papers. Here is a plot of the Impact Factor of certain journals (refers to the number of times papers are cited from that journal in a given time period) vs its retraction index. the number of retractions in a given time period relative to the number of published articles with abstracts during the same period. The prestigious journals do rather badly. https://www.nature.com/articles/nature.2014.15951

 

[GMCPhotographics]

I am occasionally tempted by both the RF 135/1.8 and the RF 85/1.2. I previously owned the EF 135/2 and the EF 85/1.2L II that I replaced with the EF 85/1.4 IS. The 'problem' is that I ended up using the 70-200/2.8 for most of those use cases. Of the two, I'd use the 135/1.8 more than the 85/1.2, based on my historical data.

I've resisted the temptation so far, and I'm hoping for a 70-150/2 to pair with the excellent 28-70/2.

The EF 135 L was a lot more user friendly and easier to get a precise focus than the EF 85 IIL. The slimmer DOF of the 85L could be challenging!

 

[SvO2]

Looks like Canon cranked up the built in raw noise reduction.

As discussed earlier, the evidence that either camera has noise reduction is weak.

There is evidence there is some sort of signal processing present which most have assumed to be NR. With the R6ii, that pattern is much more pronounced than the R5ii (in mechanical / EFCS anyway... for ES, the R6ii seems to do no signal processing). The relevant energy spectra reproduced below

 

[AlanF]

As discussed earlier, the evidence that either camera has noise reduction is weak.

There is evidence there is some sort of signal processing present which most have assumed to be NR. With the R6ii, that pattern is much more pronounced than the R5ii (in mechanical / EFCS anyway... for ES, the R6ii seems to do no signal processing). The relevant energy spectra reproduced below

View attachment 219083
View attachment 219084

View attachment 219081View attachment 219082

Could you please explain what is being plotted and how the information is interpreted. I know he says curvature shows processing, but why is that?

 

[neuroanatomist]

Scientific fraud is a serious problem. Harder data of where it occurs may be found in Retraction Watch which lists retracted papers.

Indeed (posted that link earlier). I know from corporate data that internal replication of published data was mostly unsuccessful. As I said, it’s not all due to manipulation – consider that many commonly-used cell lines are decades old and have been passed many times in different hands. Even if the line is still correct, different labs’ lines have genetically drifted.

It’s not a new problem. When I was a postdoc, the journal Biochemical and Biophysical Research Communications, abbreviated as BBRC, was colloquially referred to as Bad Biochemistry, Read Carefully.

 

[SvO2]

Could you please explain what is being plotted and how the information is interpreted. I know he says curvature shows processing, but why is that?

The technical details here go far more indepth than I can explain: https://www.photonstophotos.net/Gen...ion_to_Energy_Spectra_for_Sensor_Analysis.htm

But the simple version is that that the deviation from .5 indicates that a pixel is being influenced by its neighbors to generate the final output. NR is a common culprit of this and is a reasonable assumption in the case of photography, but theoretically all sorts of signal processing schemes can generate such a pattern.

Interestingly here Bill does note the exact pattern we're discussing and attributes it to low pass filtering and speculates it might have something to do with how the autofocus pixels are handled https://www.photonstophotos.net/Gen...lysis_Primer/Energy_Spectra_and_Filtering.htm

 

[vikingar]

Something else to think about. Nature itself has a very high proportion of retracted papers. Here is a plot of the Impact Factor of certain journals (refers to the number of times papers are cited from that journal in a given time period) vs its retraction index. the number of retractions in a given time period relative to the number of published articles with abstracts during the same period. The prestigious journals do rather badly. https://www.nature.com/articles/nature.2014.15951

There are a lot more eyeballs on the prestigious journals and conferences, and the pressure to retract gets a lot higher once a problem is made public. But of course the motivation to falsify data would be higher, as these journals publish only the exciting stuff.

At smaller journals and conferences people care a lot less. Fraud or falsification may go entirely undetected unless someone really looks. Smaller errors and inaccuracies might not even really interest anyone. It's frustrating.

Academia is a strange beast and in smaller subfields everyone knows each other and everyone wants to get ahead and publish. Morals only exist on paper, it's more like a shark tank.

I don't have personal experience with retractions, but I've heard that calling out someone or even just asking privately can be a career-limiting move. Unless there is undeniable proof of intentional fraud, practically the worst that will happen is a nice email asking the authors about it. Generally nobody will retract their paper unless absolutely forced to by a higher authority.

 

[DBounce]

The Canon sensors show just as much detail and sharpness as the Z8, A1, etc.
What is the issue here? All camera makers process (“cook”) the raw data…Otherwise you would not see an image.

The issue is, if noise reduction is already applied in camera, there would likely be little to gain by applying it in post. Whereas for other cameras their maybe additional latitude in post. With the R5MkII you are stuck with whatever Canon dictates.

 

[Quarkcharmed]

The issue is, if noise reduction is already applied in camera, there would likely be little to gain by applying it in post. Whereas for other cameras their maybe additional latitude in post. With the R5MkII you are stuck with whatever Canon dictates.

The processing Canon applies to raw files is very mild and won't prevent you from applying another noise reduction in post. Noise reduction in Lightroom works fine with R5's raw files and works fine with R5II's files.
Generally the baked-in noise reduction only affects very deep shadows.

 

[DBounce]

The A1 i guess? is the camera's competitor in sony's realm - but even then, it's in a another price bracket. I orignally didn't have the A9 III in there, , it was just put in there because if I didn't - I know some would go "where is the A9 III????!!"

I mentioned this with the Z 8 - maybe i should have added something to the A9 III - I may

You don't have the option to "turn off" the global shutter on the A9 III to use it as a general-purpose tool, that's not Canon's problem and it's still nearly $2000 more and still doesn't shoot lower than ISO 200. Again, some people shoot lower than ISO 200 - can the A9 III? well no. Canon has the ability if you need more DR to switch to mechanical shutter - can you do that with the Sony? well no. Who's problem is that? Canon's?

if you are shooting sports, etc you are mostly shooting > ISO 800 and there is no difference, with the exception of readout speed and price.

Actually you are wrong. The A9III can shoot down to ISO 125. Don’t take my word for it… check the manual.

 

[DBounce]

Looking how close R5 and R5ii are regarding DR I wonder what was the reason for R5 and R5c not having clog-2 for video. It seams the sensor has the DR.

Dynamic range in stills is different than dynamic range in video.

 

[Quarkcharmed]

NR is a common culprit of this and is a reasonable assumption in the case of photography, but theoretically all sorts of signal processing schemes can generate such a pattern.

Most likely it is noise reduction, judging by how very deep shadows look like in the R5 (they're kinda posterised). The deep shadows from the R5II, from the samples I've seen, look like more 'normal' noisy shadows.

 

[Slayer117x]

I really appreciated this article. All the DR talk had me really worried about choosing to upgrade from the 5Dmk4. I'm still worried about the autofocus and wish it had cross type on it. I know the reviews from people using it have been remarkable on focus accuracy and speed, I think at this point they are all people paid by Canon to say that, so hoping it really proves true.

 

[neuroanatomist]

Actually you are wrong. The A9III can shoot down to ISO 125. Don’t take my word for it… check the manual.

Sure, and Canon cameras can shoot down to ISO 50. But like ISO 125 on the a9III, that's an in-camera one-stop pull. The native ISO range of the a9III is 250-25600, with an expanded range (in-camera pull/push) of 125-51200...check the manual.

You can also check the manual for the R1, where the native range is 100-102400, with an expanded range of 50-409600.