Pixel Density and Reach--Any Definitive Tests?

clk_walker said:

I was hoping that the differences between the 5DII and 7D would be more significant since I just got a 7D for its "reach" advantages.

Click to expand...

They are more significant!

Look at how Alan has cropped down the 5DII image - you can do the same with the 7D image too you realise.

Think a bit more about the testing scenario chosen here before you draw conclusions.

If you can crop a 21MP image down to 8MP and not loose a great deal of resolution when compared to an 18MP image - then you can also crop down the 18MP image to 8MP with similar results.

Now can you see the reality of the reach capabilities of the two?

Alans test is not testing reach abilities at all - it's simply comparing a cropped 5DII image to a 7D image.

Come on guys! You are smarter than this!

For the comparison to be valid, the two sets of images must be resized to give the identical subject size, not the same % reduction. With the ones you've shown, all the 7D images are at larger subject magnification so we can't tell if the difference is due to the difference is sensor or difference in magnification. In addition, there is some difference is contrast (especially with the last pair of images), and contrast can effect perceived resolution/sharpness.
--
Jeff Peterman

Any insults, implied anger, bad grammar and bad spelling, are entirely unintentionalal. Sorry.
http://www.pbase.com/jeffp25
http://www.jeffp25.smugmug.com

Thanks and excellent question. I should have clarified that I was cropping the image to 12in x 8in @400 dpi. That produces an image of 4800 x 3200 pixels.

Now, the 7D's native image size is 5184 x 3456, so if I took a 7D image and cropped it by 20% for example w/o doing any upscaling/downscaling, the new image size would be 4147 x 2765.

If I take the above example and crop it by 20%, but set the crop dimensions to 12in x 8in @ 400 dpi, it the resulting image size will always be the same... 4800 x 3200. Photoshop will upscale the 7D's cropped image from 4147 x 2765 to 4800 x 3200.

I used this as part of the test, because each camera has different pixel sizes and to compare apples to apples, I resized each image to the exact same dimensions.

I feel this is a valid test, because we're always cropping/upsizing/downsizing images depending upon their final use: publishing, printing, web page, etc.

Thanks,

Alan
http://www.iwishicouldfly.com

Peter 13 said:

Great test, thank you for doing it.

Just one remark - when you say that you resize to 400DPI, I have no idea what that means. What is the uprez coefficient for the 7D image, for example?

Click to expand...

schmegg said:

clk_walker said:

I was hoping that the differences between the 5DII and 7D would be more significant since I just got a 7D for its "reach" advantages.

Click to expand...

They are more significant!

Look at how Alan has cropped down the 5DII image - you can do the same with the 7D image too you realise.

Click to expand...

He cropped all images to identical subject composition; the whole point of the test.

schmegg said:

Think a bit more about the testing scenario chosen here before you draw conclusions.

If you can crop a 21MP image down to 8MP and not loose a great deal of resolution when compared to an 18MP image - then you can also crop down the 18MP image to 8MP with similar results.

Click to expand...

That would be an interesting test; 8MP from 5DII vs. 8MP from 7D. I'll give 10:1 that the 5DII will win with very noticeable results. Alan, you up for another test?

schmegg said:

Now can you see the reality of the reach capabilities of the two?

Alans test is not testing reach abilities at all - it's simply comparing a cropped 5DII image to a 7D image.

Come on guys! You are smarter than this!

Click to expand...

Maybe I've got it wrong (as is often the case), however I think you miss the core of what Alan's test shows. Identical (cropped) composition from the same distance with the same lens, varying only the camera. He's showing the final subject detail you will realize in focal-length limited situation. This is, very effectively IMO, testing the 'reach' ability of the three cameras in a real world type of situation.

Cheers,
GT

--
Lyin' Pete used to do this all by touch...

aayjay said:

Based on feedback, I have written a 3-way comparison between the 5DM2, 7D and 1D Mark IV in a focal-length-limited test. I used the Canon 600mm f/4 IS to perform the test without a teleconverter and used various ISO settings: 100, 400, 800 and 1600.

Click to expand...

Here the first part of your conclusion:

"From a bird photographer's perspective, all three of these cameras produce similar image quality. In my opinion, the differences are so small that one should look elsewhere when attempting to decide which camera is best."

I disagree.

Unfortunately, your target subject is not very good to evaluate real subject detail because it mainly has straight lines and no fine and complex textures. However, even in your test the 7D image shows more detail, as expected. I did a small test myself and the 60D produced more subject detail than the 5D2 using the same lens from the same distance, easily visible even at ISO1600. Noise was similar, again as expected.

You state that you use the 600/4 a lot with the 1.4x TC. Do you realize that using the 5D2 with lens+1.5x TC is equivalent to using the bare lens with the 7D? If you think the 7D doesn't produce worthwhile additional subject detail, why do you use the TC with the lens? If you do a similar test comparing subject detail with and without TC you might be surprised that the differences are smaller than you apparently think.

Or another example: If you think that the subject detail produced by a 5D2 and a 7D with the same (sharp) 600mm lens is "similar", why don't you use a 400mm lens instead because subject detail will also be "similar"?

1Balance said:

Maybe I've got it wrong (as is often the case), however I think you miss the core of what Alan's test shows. Identical (cropped) composition from the same distance with the same lens, varying only the camera. He's showing the final subject detail you will realize in focal-length limited situation. This is, very effectively IMO, testing the 'reach' ability of the three cameras in a real world type of situation.

Click to expand...

Yes, Alan's test is fundamentally correct (he should not have downsampled the 7D though). Unfortunately, he comes to the wrong conclusion, perhaps because he expected bigger differences or because the lens was not very sharp or slightly misfocused or because the test subject is not ideal.

The 7D image was not downsampled. All three images were upsampled.

aayjay said:

Thanks and excellent question. I should have clarified that I was cropping the image to 12in x 8in @400 dpi. That produces an image of 4800 x 3200 pixels.

Now, the 7D's native image size is 5184 x 3456, so if I took a 7D image and cropped it by 20% for example w/o doing any upscaling/downscaling, the new image size would be 4147 x 2765.

If I take the above example and crop it by 20%, but set the crop dimensions to 12in x 8in @ 400 dpi, it the resulting image size will always be the same... 4800 x 3200. Photoshop will upscale the 7D's cropped image from 4147 x 2765 to 4800 x 3200.

Click to expand...

I'm not so sure anymore that I understand what you did. Can you explain again?

The critical thing is that you should crop the image from all cameras to the same subject framing and then up -sample all images to the same pixel dimensions.

aayjay said:

The 7D image was not downsampled. All three images were upsampled.

Click to expand...

OK. Form your (a bit convoluted) description above I concluded that the 7D's native image of 5184 x 3456 was downsampled to 4800 x 3200 pixels. Sorry.

Everyone seems to have an opinion on this subject, with opinions going in both directions. So what I was trying to do with this thread was see if we could instead look at some objective test results. With liquidstone's and aajay's help, I think we've been able to do that, and I really appreciate their taking the time to post these results.

Although I'm not sure either of these tests is entirely definitive, they've certainly shed some light on this issue from which I think I've been able to draw some conclusions. Since everyone else seems to feel the need to share their conclusions, here are mine:

1) the 7D's reach advantage is not a myth; it's real

2) the 7D's reach advantage is not night and day, but it is noticeable

3) the 7D's reach advantage only applies at lower ISOs--100-400, maybe 800

4) the 7D's reach advantage only applies in situations where the shooter is significantly focal length limited

5) the 7D reach advantage is only likely to be realized with very good (high resolving) lenses

The bottom line for me is as follows: If I were shooting critters in the wild, and I found myself significantly focal length limited shooting with my longest lens, and I had the 1DIV, 5DII and 7D all available to me, I'd shoot the critter with my 7D--at least if I could do so at 400 ISO or below. Maybe even 800 ISO. 1600 ISO not so much.

This doesn't mean that I think anyone should necessarily buy the 7D in preference to one of these other cameras. There are lots of other factors that need to be considered in such a decision. But, IMO, if you're a focal-length-limited wildlife shooter, I certainly think that the 7D's reach advantage is one of the factors you should be considering.

My 2 cents, and thanks again to liquidstone and aayjay.

Your conclusions are very reasonable.

bhollis said:

3) the 7D's reach advantage only applies at lower ISOs--100-400, maybe 800

Click to expand...

Yes, the reach advantage will diminish at high ISO, however the 7D will never be worse than the 5D2. In fact it will remain slightly better even at very high ISO, especially if good noise reduction algorithms are used.

Sorry about the confusion. All images were up-scaled/up-sized. There were pixels added to the images, but not taken away.

With all three of these cameras we are dealing with different pixel densities and crop factors and the only way I know how to make a fair comparison is to upsize them all to the same dimensions.

In their native (crop) for the test, the three cameras cropped to the same field of view would produce this (7D, Mark IV, 5DM2):



As you can see in order to equal their sizes, the 5DM2 required more up-scaling than the Mark IV and the Mark IV required more up-scaling than the 7D.

Of course, the more up-scaling you do the more it will degrade an image, but we do it all the time when sending images to a high-resolution printer for example. If it were not upscaled, the print-out would not fill the intended page.

It is truly amazing however that these images can be upscaled to huge dimensions. I've been working on a project where some of my images are going to be wall-size. They really look spectacular, but of course they are being viewed from a distance.

Alan
http://www.iwishicouldfly.com

mattr said:

aayjay said:

Thanks and excellent question. I should have clarified that I was cropping the image to 12in x 8in @400 dpi. That produces an image of 4800 x 3200 pixels.

Now, the 7D's native image size is 5184 x 3456, so if I took a 7D image and cropped it by 20% for example w/o doing any upscaling/downscaling, the new image size would be 4147 x 2765.

If I take the above example and crop it by 20%, but set the crop dimensions to 12in x 8in @ 400 dpi, it the resulting image size will always be the same... 4800 x 3200. Photoshop will upscale the 7D's cropped image from 4147 x 2765 to 4800 x 3200.

Click to expand...

I'm not so sure anymore that I understand what you did. Can you explain again?

The critical thing is that you should crop the image from all cameras to the same subject framing and then up -sample all images to the same pixel dimensions.

Click to expand...

Yes, I agree here. Even at higher ISO's there are differences. I didn't include noise reduction and sharpening in the test, but after applying NR and sharpening, there still are differences.

Here's how they look after applying NR and sharpening at ISO 1600:



Other than that, the OP is bang-on.

Fun Stuff!

Alan
http://www.iwishicouldfly.com

mattr said:

bhollis said:

3) the 7D's reach advantage only applies at lower ISOs--100-400, maybe 800

Click to expand...

Yes, the reach advantage will diminish at high ISO, however the 7D will never be worse than the 5D2. In fact it will remain slightly better even at very high ISO, especially if good noise reduction algorithms are used.

Click to expand...

Impressive showing for the 7D! What NR & sharpening?

Do the 7D images take to sharpening better than 5DII? I.e. would the 7D's pixel-density reach be more evident in fully processed images?

Cheers,
GT

--
Lyin' Pete used to do this all by touch...

1Balance said:

Impressive showing for the 7D! What NR & sharpening?

Click to expand...

Topaz Labs DeNoise 5 for NR and CS5 Unsharp Mask.

1Balance said:

Do the 7D images take to sharpening better than 5DII? I.e. would the 7D's pixel-density reach be more evident in fully processed images?

Click to expand...

I'm not sure if it takes to sharpening better or not. I guess you could claim that since there is more detail in the 7D image. From a Noise Level perspective, when the images from all three are cropped identically, their noise levels are quite similar.

And also remember we are looking at these images at the pixel level. The differences are there, but when looking at an entire processed image, it would really be difficult to tell these three images apart.

Alan
http://www.iwishicouldfly.com

1Balance said:

Cheers,
GT

--
Lyin' Pete used to do this all by touch...

Click to expand...

mattr said:

1Balance said:

Maybe I've got it wrong (as is often the case), however I think you miss the core of what Alan's test shows. Identical (cropped) composition from the same distance with the same lens, varying only the camera. He's showing the final subject detail you will realize in focal-length limited situation. This is, very effectively IMO, testing the 'reach' ability of the three cameras in a real world type of situation.

Click to expand...

Yes, Alan's test is fundamentally correct (he should not have downsampled the 7D though). Unfortunately, he comes to the wrong conclusion, perhaps because he expected bigger differences or because the lens was not very sharp or slightly misfocused or because the test subject is not ideal.

Click to expand...

Frankly I think you have got to the point of being very boring. Read what Alan did and why. His comparisons are exactly the type of test I ahve been wanting every since the 7D came out. Yes there is a supposed "reach" advantage to the 7D and I was hoping that it would be very significant. But his test results and others like it do not show it.

Resolution is certainly higher for the 7D. But that is looking at MTF values of 5-10% on a high contrast subject. Not many of those in bird land. In fact bird feathers tend to be relative low contrast and varies alot with the sun angle.

The optical reason that one uses a TC is to project a larger image on the image plane of the camera. This basically decreases the LP/mm of the subject features and puts those features in a more desirable place on the sensor MTF curve. If the subject had a feature at say 100LP/mm and the sensor cutoff was around this value the sensor response would be very low. Adding a 2x TC transforms the 100LP/mm feature to a 50LP/mm feature would would be significantly above the sensor cutoff. Hopefully this explains this to you.

In my case I tend to shoot alot of birds at my bird feeder at the MFD of the 400mm f/5.6 lens. I have used a 20D since 11/2005. Looking at the 7D seemed promising but I wanted some real life tests. Alan's tests results are somewhat dishearting but I find no fault in his test results or method. Others have done tests with paper notes which show some differences. I had a very hard time seeing the differences in Romy's- test but others did.-
A bird in the viewfinder is worth...

One other consideration that must be accounted for is camera and/or subject motion. If you are using the 5DII at minimum shutter duration for the given scenario then the 1DIV and 7D would have to have shorter shutter durations to compensate. This would lower the SNR some or if not compensated their images would be blurred somewhat.

Another consideration concerning maximum resolution: This only occurs at the plane of best focus (PBF). On either side of the PBF, out to the limits of the DOF, the resolution will be degraded. Birders will typically try to focus on the bird's eye. If not then image resolution will be degraded. Increasing the F# to increase the DOF will have more of a negative impact on the 7D than the 5DII due to diffraction.

All of these are real world factors that are difficult to control in a test scenario. I remember reading in this forum of an attempt at such a test by Peacefrog. Boy did he get castigated!!!
--
A bird in the viewfinder is worth...

aayjay said:

Thanks and excellent question. I should have clarified that I was cropping the image to 12in x 8in @400 dpi. That produces an image of 4800 x 3200 pixels.

Now, the 7D's native image size is 5184 x 3456, so if I took a 7D image and cropped it by 20% for example w/o doing any upscaling/downscaling, the new image size would be 4147 x 2765.

If I take the above example and crop it by 20%, but set the crop dimensions to 12in x 8in @ 400 dpi, it the resulting image size will always be the same... 4800 x 3200. Photoshop will upscale the 7D's cropped image from 4147 x 2765 to 4800 x 3200.

I used this as part of the test, because each camera has different pixel sizes and to compare apples to apples, I resized each image to the exact same dimensions.

I feel this is a valid test, because we're always cropping/upsizing/downsizing images depending upon their final use: publishing, printing, web page, etc.

Click to expand...

Thanks. Yes, it is. I am not questioning the validity of the test. It is rare to see here well done tests, and this is one of those.

clk_walker said:

...Adding a 2x TC transforms the 100LP/mm feature to a 50LP/mm feature would would be significantly above the sensor cutoff. Hopefully this explains this to you.

Click to expand...

Adding a 2xTC will do the same thing as using a sensor with 1/2 the linear pixel size, as far as the resulting image is concerned. Hopefully this explains it to you.

clk_walker636320 said:

I had a very hard time seeing the differences in Romy's- test but others did.-

Click to expand...

Romy's test shows well what you can expect when using a sharp lens and good technique. I think the increased subject resolution with the 7D is very significant in his test. If you don't see this difference, a 7D or a TC for your lenses might not be a wise equipment acquisition (if the goal is increased reach).

aayjay said:

Sorry about the confusion. All images were up-scaled/up-sized. There were pixels added to the images, but not taken away.

Click to expand...

Yes, this is how it should be done. IMO even the 7D crop (which has the largest pixel dimensions) should be up-sampled a bit in order to subject all images to some re-sampling (Romy didn't do that in his test; I'm not sure it makes a difference).