To crop or to use an extender with the 5DII?

[Feanor]

I'm thinking of buying myself the 400 f5.6 prime for Christmas. I already have a 2x TC but it'd reduce the aperture too much on the 400, and I wouldn't be able to afford the 1.4x TC for a while.


But that got me thinking. As far as I understand it, a teleconverter just magnifies the existing image, right? And you lose a stop or two of light because the light coming in from the front of the lens is, effectively, being spread more thinly over the sensor?


Surely, assuming that loss of resolution isn't an issue (for argument, using an ∞MP camera), the image degredation (CA etc) will be the same as cropping? In fact I'd expect a cropped image to look better, as it's not being sent through any extra glass.


In my experience with using my zooms on my 5DII, image degredation becomes an issue with cropping long before loss of resolution does; and besides, a 1.4x crop is still using more of the sensor than an APS-C camera. I realise that resolution is more likely to become the limiting factor for cropping with primes, but the 5DII has so much to play with that even then I'm not convinced it will be an issue for day-to-day print sizes.


What's the opinion of the experts? On today's modern high-res sensors, should we be using TCs at all?

[Colin]

In my non-professional experience, unless you're actually dealing with a pixel resolution issue (which you can in some cases, though I don't know about a 5D2... teleconverters are usually worse than just cropping, but it will depend a little on the lens too. Some lenses seem to be more friendly to a teleconverter. If the pixel resolution doesn't make full use of the lens, it may very well be better to use a teleconverter.


Is that ambiguous enough?

[Daniel Browning]

As far as I understand it, a teleconverter just magnifies the existing image, right?


Yes.



And you lose a stop or two of light because the light coming in from the front of the lens is, effectively, being spread more thinly over the sensor?


No. The amount of light lost from the TC is the exact same as the amount of light lost from cropping. Furthermore, the amount of light from the *subject* is the exact same. In other words: as it pertains to the subject, you have less light intensity, but more area, so the total amount of light from the subject is the same.






Surely, assuming that loss of resolution isn't an issue (for argument, using an ∞MP camera), the image degredation (CA etc) will be the same as cropping?





With an ∞MP camera, the TC would have more aberrations than cropping..



On today's modern high-res sensors, should we be using TCs at all?


Yes. If you think 21 MP on FF is enough to make TC unnecessary, then that means you also believe 8MP on APS-C is enough to make TC unnecessary. I find TC highly effective on 8 MP APS-C, even with zooms.


The other factor to keep in mind is noise. For example, try this comparison:


5D2, ISO 100, f/2.8 (cropped digitally by 2X), 1/1000


5D2, ISO 400, f/2.8+2X TC (f/5.6), 1/1000


You will find that the second shot has far, far less noise.

[Feanor]

In my non-professional experience, unless you're actually dealing with a pixel resolution issue (which you can in some cases, though I don't know about a 5D2... teleconverters are usually worse than just cropping, but it will depend a little on the lens too. Some lenses seem to be more friendly to a teleconverter. If the pixel resolution doesn't make full use of the lens, it may very well be better to use a teleconverter.


Is that ambiguous enough?
<div style="clear: both;"]</div>





I think you're pretty much saying what I was trying to :).


On older bodies, there'd have been reason to use a TC. Back in the days of the 30D when you were looking at 8MP sensors then pixelation would probably have been an issue before light degredation. But on a 21MP FF camera like the 5D2 I strongly suspect it's either roughly 50-50 or the other way round, possibly by a significant margin.

[BCalkins]

I think it depends on a number of factors. I just sold my 1.4x converter because I just didn't find it gained me anything in the types of shots I was taking. When I did need it (because somethat was really far away) it wasn't enough. To view on screen or print small prints I don't think it matters (and cropping is cheaper).


The converter has the advantage that you can frame more precisely... In the meantime, it is free to shoot and crop and see if you are happy with the results. If not - then you are really deciding between the 1.4/2x or the longer lens.

[Daniel Browning]

Back in the days of the 30D when you were looking at 8MP sensors then pixelation would probably have been an issue before light degredation. But on a 21MP FF camera like the 5D2 I strongly suspect it's either roughly 50-50 or the other way round, possibly by a significant margin.


But the 8MP on 30D is *exactly* the same as 21MP on the 5D2. If the lens is good enough to use TC on the 30D, then it is good enough to use TC on the 5D2.

[Jon Ruyle]

On today's modern high-res sensors, should we be using TCs at all?


Yes. If you think 21 MP on FF is enough to make TC unnecessary, then that means you also believe 8MP on APS-C is enough to make TC unnecessary. I find TC highly effective on 8 MP APS-C, even with zooms.





Daniel is right- there are many people who want the MP's. However the question you should be asking isn't "should we" but "should I". If 5mp is enough for you, I say forget the teleconverter and crop.


I don't tend to get good enough results with a 2x teleconverter to mind losing the mps, so I rarely use it. But I use mine mostly with a zoom lens... maybe with high quality primes it is a different story.






The other factor to keep in mind is noise. For example, try this comparison:


5D2, ISO 100, f/2.8 (cropped digitally by 2X), 1/1000


5D2, ISO 400, f/2.8+2X TC (f/5.6), 1/1000


You will find that the second shot has far, far less noise.





I don't think so, Daniel. They'll be the same- if anything, the second will be a little noisier due to the light loss from the extra glass of the teleconverter.


Of course it is possible that I'm confused, in which case I don't doubt you'll explain to me the error of my ways [:)]


One thing I often wonder about is how many abberations are introduced by the teleconverter. Teleconverters have a nasty reputation for introducing optical problems, but my guess is they mostly just magnify those that are there already. Anyhone done a test of this? What I have in mind is a comparison of a cropped image with a teleconverter image done in a way that does not display limits of sensor resolution.

[Daniel Browning]

They'll be the same- if anything, the second will be a little noisier due to the light loss from the extra glass of the teleconverter.


You are right if we only consider photon shot noise, but when it comes to read noise, the TC will be superior. For wildlife, I'm always hitting read noise because of the fast shutter speeds and long lenses required.






One thing I often wonder about is how many abberations are introduced by the teleconverter. Teleconverters have a nasty reputation for introducing optical problems, but my guess is they mostly just magnify those that are there already.





I think you're right: a lot of their bad reputation comes from simply magnifying existing aberrations in the underlying lens. Thankfully, the worst part of the aberrations are some C.A. and distortion that are easy to correct.






Anyone done a test of this?





I'm sure there are some good tests out there somewhere for this, but I can't recall any. It would have to be certain that the pixel size is small enough to completely oversample the lens, so as technology marches on this test will get better with more and more lenses.

[Jon Ruyle]

You are right if we only consider photon shot noise, but when it comes to read noise, the TC will be superior. For wildlife, I'm always hitting read noise because of the fast shutter speeds and long lenses required.


You're right- I was thinking of photon noise only. Do you get quantization noise if you shoot lower iso and underexpose? (You must I guess... or you would.)

[Daniel Browning]

Do you get quantization noise if you shoot lower iso and underexpose? (You must I guess... or you would.)


Never. The read noise is always so high that it completely suppresses quantization noise (AKA quantization error). In order to get quantization error, you would have to reduce read noise down to amazingly low levels without increasing bit depth. In practice, no one does that, because bits are cheap. In fact, in real life we have the *reverse* problem: camera manufacturers adding way more bits than necessary, when there has been no improvement in pixel-level read noise.

[Feanor]

Some very interesting points, thanks :).







Back in the days of the 30D when you were looking at 8MP sensors then pixelation would probably have been an issue before light degredation. But on a 21MP FF camera like the 5D2 I strongly suspect it's either roughly 50-50 or the other way round, possibly by a significant margin.


But the 8MP on 30D is *exactly* the same as 21MP on the 5D2. If the lens is good enough to use TC on the 30D, then it is good enough to use TC on the 5D2.
<div style="clear: both;"]</div>





I must have started writing my reply when you posted your original reply, I only just noticed it, sorry. I don't understand though what you mean by the 8MP on hte 30D being the same as the 21MP on the 5D2? 8MP x 1.6 gives 12.8MP, so the 5DII has another 1.6 times the pixel density as the 30D by my rough calculations. Surely that means you can make a 1.6x crop from the 5D2 twice over and still get the same resolution as the 30D? (one of the crops taking you from FF to APS-C equivalent)


And I see your point about ISO100 vs ISO400, but what about ISO800 vs ISO3200 for example? Surely at that threshold the read noise increase will be greater?

[wickerprints]

A 24x36mm "full frame" sensor is 1.6 times larger in linear dimension than a 22.3x14.9mm APS-C sensor. In terms of sensor area, the full frame has 1.6 * 1.6 = 2.56 times the area of the APS-C sensor.


Since pixel count is based on the area of a sensor and not its linear width/height, the appropriate math is


21 MP / 2.56 = 8.2 MP,


so Daniel quoted the correct value.

[Jon Ruyle]

Never.


I see. But in any case, I was being stupid. To reduce read noise, I think, you would have to use a *higher* iso, not a lower one. The problem would be blown highlights, not quantization. (Or it could be just that I'm confused again... that happens quite often).

[Daniel Browning]

And I see your point about ISO100 vs ISO400,
but what about ISO800 vs ISO3200 for example? Surely at that threshold
the read noise increase will be greater?


At ISO 3200 vs. 800, the comparison will still favor the TC. Furthermore, the difference will be visible in midtones instead of just the deep shadows. This is one test that anyone with a TC can do.



To reduce read noise, I think, you would have to use a *higher* iso, not a lower one. The problem would be blown highlights, not quantization.


Agreed.

[Feanor]

A 24x36mm "full frame" sensor is 1.6 times larger in linear dimension than a 22.3x14.9mm APS-C sensor. In terms of sensor area, the full frame has 1.6 * 1.6 = 2.56 times the area of the APS-C sensor.


Since pixel count is based on the area of a sensor and not its linear width/height, the appropriate math is


21 MP / 2.56 = 8.2 MP,


so Daniel quoted the correct value.
<div style="clear: both;"]</div>





Ah OK that makes sense, thanks :). Also I checked the specs of the 30D ("http://www.the-digital-picture.com/Reviews/Canon-EOS-30D-Digital-SLR-Camera-Review.aspx) and of course both you and Daniel are right about the pixel density being the same. Which surprises me - why didn't we have a 21MP full-frame sensor in the days of the 30D if the technology was there?? [:S] I guess it's not all about pixel density but also about processing large images quickly as well as storing them. It's an interesting point though - it means there is more to the idea of crop sensors giving greater reach than some people claim.


Daniel - thanks for your explanations too, I see what you're getting at. But if we then take the 7D as an example, with its considerably higher pixel density, then I take it that the gap between cropping and extenders is at least narrowed?

[Daniel Browning]

why didn't we have a 21MP full-frame sensor in the days of the 30D if the technology was there??


Frame rate. The more pixels you have, the slower the readout (for a given cost and technology). On top of that, the DIGIC processing would slow way down too. So they could have made the 5D Classic with 21 MP, but it would have been like 0.5 FPS instead of 3 FPS. That's too slow for most of the market, and it would have sold even worse than it did.






But if we then take the 7D as an example, with its considerably
higher pixel density, then I take it that the gap between cropping and
extenders is at least narrowed?





Yes. The 7D has 49% higher linear resolution than the 30D, so there are some lenses that no longer benefit from a 1.4X TC, particularly the lowest quality zooms. (Good zooms, like the 70-200, still benefit from TC on the 7D.)

[Feanor]

Yes. The 7D has 49% higher linear resolution than the 30D, so there are some lenses that no longer benefit from a 1.4X TC, particularly the lowest quality zooms. (Good zooms, like the 70-200, still benefit from TC on the 7D.)
<div style="clear: both;"]</div>





Hmm. You've got me wondering again whether the 5DII is the right camera for my interests - after the mass bemusement in a thread suggesting I swap a while ago I shelved the idea but I'm really not that interested in landscape photography. The AF points and extra reach would really be a big boon. But then again shots that I do manage to take on the 5DII with the 400mm - assuming I can get close enough and assuming the AF points don't let me down - promise to be stunning.