I have uploaded a mini review of the 1Ds3 with on my workshop site, complete with comparative crops from the 1Ds3, 1Ds2 and 5D for those interested.

Click the thumb to view:



Cheers,

Hi Jack,

Very interesting results - that building looks fine for doing these kind of tests. I agree with your visual conclusions from what I could see on my display (LaCie 321). However, there is one aspect I must ask you about that may influence the results: it looks to me, so I'm asking you - do you notice that the subject is very slightly darker in the 1Ds3 sample than in the other two? Did you have them set on Aperture Priority, and is it possible that the shutter speeds varied slightly between the three cameras? The reason I ask is that I'm wondering whether part of the apparent resolution advantage of the 1DsIII may be due to slightly more definition arising from a slightly darker exposure?

Very informative, Jack.

I've been delighted with my 5D -- until now. I guess I'll have to post the current prices of the 1Ds3 and 5D above my monitor as a constant reminder that my 5D is adequate for my purposes.

Hi Mark:

I actually adressed that in the article. ALL images were shot at the exact same shutter speeds at each ISO and the histo confirms they are essentially identical. If you look at the brighter areas you can confirm this. They were all also converted with identical settings save for detail and sharpening. However, bumping the fill slider to 30 in each image had a stronger (but nearly identical) effect on the 5D and 1Ds2 image than it did on the 1Ds3. For some reason the 1Ds3 did not respond to the fill slider at the same rate and I don't have an explanation for that.

I point you to next to the gray unpainted vertical strip of wood in the center and the white paint at the right --- these are very similar in all images, yet the 1Ds3 clearly resolves more grain in the gray wood. Red seems to render differently in the 1Ds3, a bit darker, and this could explain the slightly darker tiles and different coloration of the brown vertical piece of wood and tan roof shingles. I think the sand in the roof shingles has enough contrast to render even if they are a bit brighter in the 5D, so I really feel the granularity is probably beyond the capabilities of that sensor to resolve it.

Again, I think more complete testing would need to be done to make a conclusive determination.

Cheers,

Removing the AA filter is $450, to improve a $8000 camera it might make $ense
Marc

Not sure why anyone would want to do that.

The whole system is built around a sensor with AA filter, including third party raw converters, and I do not believe that any of those treat a file independantly from the expectations they have in terms of the sensor behaviour. This needs to be confirmed through testing obviously.

I am not even talking about the vignetting you are likely to encounter with an AA filter less body when shooting with wides.

If you find 1ds3 images to be too soft when viewed at 100%, my advice would be to look at another system altogether.

Cheers,
Bernard

Marc: I have a challenge for you: buy the camera, find someone expert enough to remove the filter properly (I understand from an article I read somewhere - I forget - one layer of it is glued to the sensor itself), test it for all the conditions in which the filter is either meant to assist or can be an interference, and let us know the results. You will have either blown 8500 bucks or opened a significant curtain for the digital imaging community!

Jack - it was late at night and I probably missed that sentence. But OK - I see you confirm parts of the 1DsIII image render a bit darker, so it wasn't my imagination - good. You have clarified the probable cause. I agree completely with your observations about the superior resolving of grain in the wood and sand in the roof shingles - I looked long and hard at those very places last night. My only concern was to understand why - whether it is due to superior resolution or better edge contrast due to the slightly different rendering of the luminosity. Perhaps it is both.

Cheers,

Mark

Can't remember where, maybe Northlight? he showed some 1DS3 images that, notwithstanding the AA showed traces of moire. His comment was - Why would anyone want to take the filter off? I think with appropriate capture sharpening I'm seeing an image that's acceptably sharp. The risk of image degradation without the AA is something that I have no motivation to assume. As an aside, I'm not sure what effect the anti dust mechanism might have on a retrofit, a la the Maxmax mod to the 5d.

There appears to be a slight conundrum here. It is claimed the resolution differences between the 5D and 1Ds2 are smaller than expected because the 1Ds2 has a stronger AA filter.

When I look at Jack's crops, I get a sense that the increased detail in the 1Ds3 image, compared to the 1Ds2 crop, is greater than the increase in detail of the 1Ds2 image compared with the 5D crop.

The obvious explanation is that the 1Ds3, like the 5D, does not have a strong AA filter. Yet Jack makes the observation that it does have a strong AA filter, stronger than that of the 1Ds2. Has the 1Ds3 image received more sharpening?

On the subject of AA filters, why should a higher megapixel camera have a greater need for a strong AA filter than a lower megapixel camera? I thought the idea was, greater pixel density reduced the need for an AA filter. At some point, given sufficient pixel density, there should be no need at all for an AA filter.

I see the same thing.


Normally I'd agree with this conclusion as well...

However, as I indicated in the article, I did apply a *lot* more sharpening to the 1Ds3 file in the raw converter than either the 5D or 1Ds2: ACR amount 50/Detail 50 as opposed to 30/25 for the 5D and 35/30 for the 1Ds2. That is the reason I infer the 1Ds3 has a stronger AA filter. Note however, I am only theorizing it has a stronger AA filter due to the additional capture sharpening required and I don't know that for a fact...

Jack,

Have you experimented with the Highlight Tone Priority (Custom Function II-3) setting yet? I know there is another thread discussing it, but I raise it here for your attention. I had it enabled and took some shots of dark buildings this afternoon. I found them a bit noisier than I was expecting from this camera. So I disabled it and tomorrow I shall try to get out and re-shoot the same scenes, see if it improves. Meanwhile I would really appreciate any real-world comparative experience reports (your's or anyone else reading this) of shots with and without this feature enabled.

Mark

Hi Mark:

I had it OFF for my tests, and did not have enough time to specifically experiment with it. I am interested in your results.

One thing I did notice one thing that was interesting: I took an image at ISO 1600 -1 ev and pushed it +1 in the raw converter for an effective ISO 3200 shot. The interesting part was no additional detail or depth was gained in the shadows; I ended up with a slightly higher contrast and noisier image was all. Comparing 1600 to 800, I gained a bit more depth using 1600, but probably only 1/3 to 1/2 stop total, which is why I mentioned I feel ISO 800 is about the maximum practical for this camera.

Cheers,

Great test, thanks for posting.

I see! I'm no expert on sharpening routines and those lengthy tomes on the subject, by Bruce Frazer and others, are not high on my list of reading priorities. However, is it not a fact that larger files generally either require more sharpening or can take more sharpening before they begin to look unnatural and over-sharpened?

I recall one of your own sharpening routines is to interpolate a file to a larger size, apply a greater degree of sharpening than you would normally at the smaller size and then downsample the image to its original size using 'bicubic sharper'.

That the 1Ds3 image should require greater sharpening to bring out the additional fine detail it is able to capture does not seem odd to me, or indicative of a stronger AA filter, but what one would expect when using the same lens that was used on the other cameras.

It has to be the case that every time we get an increase in pixel count, the additional pixels are capturing softer detail, less contrasty detail, if we are using the same lenses.

If, for example, the cut-off point for resolution with the 1Ds2 is 50 lp/mm and the cut-off point for the 1Ds3 is 60 lp/mm (let's not quibble about the precise numbers), we can't expect detail at 60 lp/mm (on the 1Ds3) to have the same contrast as detail at 50 lp/mm on the 1Ds2. This point should be obvious to anyone who understands MTF charts.

The reason I mention this is because already, as a result of your suggestion that maybe the AA filter on the 1Ds3 is even stronger than that on the 1Ds2, we have comments from readers on this forum that they have suddenly lost interest in the 1Ds3 and will cancel their order, or not place an order.

You've put the wind up some folks, Jack .

Sure Ray... Let's use your figures of 50 and 60 LPMM respectively --- that's a net resolution difference of approximately 20% which would imply the need to apply approximately 20% stronger sharpening to gain the same result. However in the case of the 1Ds3, I needed to apply closer to 70% more sharpening than the 1Ds2 to get to the same result. A bit of a gap to be sure, so clearly some other factor is in play; feel free to offer your own theory as to what it might be...

As to your wind up the skirts comment, I simply reported what I did to process the files that generated the crops I posted from each camera... Folks are free to draw their own conclusions.

Cheers,

Did you try any other lenses to be sure that might not be some kind of factor? I know the 135/2L has a good rep (I have one too) but if that copy had a very minor issue, it could account for the results you've gotten, and why they differ from Michael's...

Well, to be more precise, Jack, 43% more sharpening and 67% more detail enhancement. But perhaps the relationship between percentage increase in resolution and required percentage increase in sharpening is not so direct.

I think ACR has always had a default sharpening of 25%. By your reasoning, if that 25% default sharpening is adequate for a 3mp D30 image, then a 10mp 40D would require a default sharpening of something like 45%. Is this how it works? As I said, I'm no expert on sharpening routines. I just use my eyeballs (and the sliders of course ).

My figures of 50 and 60lp/mm for the 1Ds2&3 respectively were just to illustrate the concept of diminishing returns when one increases sensor pixel count without increasing lens resolution.

Having rechecked the figures at dpreview, I see that Phil Askey has made the common mistake of equating percentage increase in pixel count with percentage increase in resolution, where he states that the increased pixel count of the 1Ds3 represents a 26% increase in resolution over the 1Ds2.

In fact, of course, the increase in resolution is a mere 12.5%. We should not forget this otherwise we are likely to have unrealistic expectations of what to expect with the 1Ds3. The fact that you've been able to demonstrate that this 12.5% increase should be noticeable in a large print, using the same lens, makes me think the AA filter on the 1Ds3 is actually no stronger than that on the 1Ds2.

A year ago I had my 5D modified by MaxMax and have never regretted having the AA filter removed (for landscapes/nature) No problem with the sharpening plugins you just use a lower setting with better results. My point is every one complains about Canons strong AA filters but if it is an issue you can easily have it removed and a visible light bandpass filter installed. Not a big deal, BTW all the raw converters/plugins handle the files fine. If you like the MFDB look (no AA filter) you can have that look in the Canon for $450. So with a years positive experience with a modified 5D I don’t see it as a major issue. You want a sharper raw file, modify your camera, you want a soft raw file don’t modify your camera, simple. After sharpening the difference is small but noticeable, a small improvement. I now have 1 week of experience with a P30 and still prefer the look of a sensor without a AA filter. The modification is reversible I have the original AA filter/bandpass filter in the canon box if needed. As far as testing it for moiré yes some times it is apparent and with a little Gaussian blur (around a pixel) you corrected the occasional problem. If you were always shooting fashion of course you wouldn't consider the mod.
Marc

Sorry to post offtopic, but do you have a link to some of your pics without the AA filter? Especially ones that produced moiré. I am seriously considering doing this to my 1Ds MKIII.

Marc, very interesting - this is the first actual user report I've seen of the results from doing that. I was on the Maxmax website some weeks ago and noticed that while they offer this service for the 5D, they do not offer it for the *1* series cameras. I wonder why. Perhaps they have not invested in one of these expensive cameras to actually try it - there are build differences between the models. Before they offer this service on a 1Ds MKIII they would need to buy one and take the risk of trying to remove and replace it. Could be interesting to get in touch with them and ask about it.

I have no basis to question your observation about the difference in the results with versus with the AA filter, and it stands to reason; but one also needs to ask, in light of the comments Ray made about the relationship between higher resolution and lower acutance, whether the AA filter is the critical determinant of the need for heavier sharpening with the MkIII.

Marc:

I've been tempted for the past few months to send my 5D off and have MaxMax hotrod it, though was waiting for the opportunity to do the 1Ds3 review first

Did you by any chance do before and after comparative shots and can you comment more specifically on the improvement? Moire isn't a big deal for me, but detail is. The image they show on their website was not done with a great lens and the gain appears fairly marginal --- visible increase in micro-detail, but again, I'm not sure how that will translate into a print. If better glass (I shoot mostly Canon primes) offered a better result, I'd be all over the conversion. Again, would really like to hear more comment from you on the differences you're seeing in your images.

Thanks,

PS: As respects 1 series mods, I don't think they can be done because of the way the AA/IR filter is sandwiched to the sensor.

Jack, yes - that corroborates what I mentioned in post #7 above, but it still may be worthwhile confirming with MaxMax.

Jack - I disabled Function II-3 [HTP], will test it as such, and report the results. It's a very grey day here today, but I may still take the train downtown and re-shoot that same dark building to see whether there is less noise with that feature off. I'll let you know once I have a useful result.

Mark

I didn't have the presence of mind to take a before and after test shot so my comparisons are all subjective
Marc

LOLOL! Okay, I commend you on your honesty! Can you provide some more detailed subjective input about how the files look? I'm trying to decide if it's worth doing and I trust your brand of subjective

Thanks!

The final images look more real as if the subject is more 3 dimensional. The micro detail is better like stones sand etc look more realistic. My wife is an artist and noticed a difference without knowing the camera was modified "how come the print looks so sharp" again very subjective. I'm on the road and can't post a good example but perhaps I can u-send-it a raw for you to look at.
Marc

Jack
Here is a raw from last month (part of a stitched panorama) about 12,000' coming down Mouna Kea. Had to get my snow fix (I miss New hampshire and the snow this time of the year)
Marc

http://download.yousendit.com/C301AE1B06BAB243

It really is funny that we are once again at thet topic. As often discussed with good Captrue sharpening the difference is getting so small, that it doesn't really make sense to remove any AA Filter.
Take to pictures side by side one with AA and one without Filter. Sharpen them both for print. Compare them on screen and make a print. 99,9% of all people will NOT see ANY difference AS LONG AS the sharpening is done right.

I have seen quite a few reddot 5D sample shots, bu I have never seen one which is better than with AA filter.

Sorry but as I see it it is just making money on the believes of people.

Thanks Marc! Downloading it now!

One of the things I hope to gain is some of the inter-pixel "sparkle" that most AA-filterless cameras tend to show more of...

Edit: Looks darn good straight off the camera, and doubly impressive given it's from the 24-105 zoom at 24. It shows some light moire in the high-frequency detail area of the roadside gravel. It's not visible in a properly sharpened print however, and the rest of the image looks so good it may be worth living with that. Hmmm...

Are you surprized to see more noise if you expose one step lower than you otherwise would do? (Of course, this depends on how good the "otherwise exposure" would have been.)

Yes - based on the claims being made for this camera's noise performance; but the manual does warn us that using this feature will increase noise. This is correct.

Jack,

Further to my other message on the subject of the tests with and without HTP, I did get back downtown this morning to re-shoot images similar to yesterday's of the same group of buildings - these are tall black office towers - Torontonians will know them - the TD Centre. I had two incidents of being stopped by security asking if I was a professional photographer and what I would be doing with the pictures. The second one explained he stopped me because he noticed the camera looked kind of professional. So I told him nah - it's just an expensive P&S that allows some adjustments . He didn't buy that, wanted to know what I was doing with it, so I told him I was testing the camera for digital noise in the 3/4 tones with versus without various adjustments. That sailed straight through like s..t through a goose - and he invited me to go to the 38th floor for a camera permit. By then I was finished the job so no permit needed, but for those locals wanting to photograph on the grounds of the TD Centre with an expensive P&S - go to the 38th for a permit first - will save harassment.

OK back to the pictures. Definite improvement at ISO 200 - the shots with HTP disabled definitely display less noise. I also did a bunch, HTP disabled, at ISO in the range of 400~640 and I'm less impressed. Noise readings in Noise Ninja are in the 14~23 range (mainly high frequency Luminosity noise), most visible in skies and skin tones. I had expected better, and wonder whether this is typical for this camera, or I should take it back to Canon for testing.

If the scenery's dynamic range is not not stretching from very dark to clipping the highlights, then there is no point of using HTP; the exposure has to be decreased in case of clipping.

However, if the image contained already very dark areas, and now the exposure gets reduced by a stop, then it is only natural, that there will be some noise. Otherwise the camera could claim that the DR is one stop higher than its predecessors' (Canon did not make such claim.)

I wrote above "the exposure gets reduced by a stop"; this is not accurate, the exposure remains as indicated in the camera, the ISO will be halved. However, one could see it this way: had the effectively applied ISO been selected, the exposure should have been one stop higher.

Sounds like a feature only useful for complete novices. If you know you're in a situation where the normal autoexposure is likely to clip highlights, then you could just as easily set the camera to -1 stop EV.

Or perhaps it really does serve a purpose when shooting in jpeg mode. In other words, HTP for jpeg output is better than a one stop underexposed shot for jpeg output.

Not sure how many complete novices buy $8000 DSLRs. Maybe Canon figured it's useful for photographers who don't mind some shadow noise but appreciate the extra headroom where most of the levels will be.

Mark,
As I understand it, this feature is not available at ISO 100 (or is that ISO 50 on the 1Ds3) because the camera automatically lowers the ISO one stop without telling you. Does anyone except a novice want the camera to make such decisions without any prior analysis of the scene. An automatic exposure results from an assessment of the lighting conditions by the camera's metering system. This HTP feature, however, seems to be no different than the user setting the camera to -1EV. In both cases a button has to be pressed or a wheel turned.

I'm struggling to find a purpose for this feature.

1. The in-camera image display is one stop brighter than it would be if you lowered the exposure or the ISO.

2. When recording JPEG (or raw and JPEG) format, the exposure is "corrected" and the highlight contrast is reduced somewhat in order to pull back the right end. The same is true for the JPEG embedded in the raw file, but I have not seen anyone, who made use of the that almost-halfsized JPEG image.

Okay! It sound as though this feature is only useful for those who shoot in jpeg mode. There have been times, long ago when memory cards were expensive, that I switched from RAW to jpeg as my last card filled up. As I recall, despite being careful with exposure, there were shots with blown highlights that I could do nothing about.

This HTP feature seems like an insurance policy for those who shoot jpeg.

Those recording "raw only" have no reason to use HTP. The brighter in-camera display is not really important IMO, and the one full stop reduction is too rigid; one can reduce the exposure by 1/3 or 2/3 to avoid clipping. (Of course, onle can use HTP with +1/3EV or +2/3 EV bias to achieve the same.)

HTP does not help on the main problem: that one does not know if clipping occured (not even in post processing, because not only the camera does not show true raw histogram, but the raw processors are lying about clipping).

However, there are people, who shoot raw+jpeg, with the intention to use the jpeg if it looks ok, and work on the raw only in exceptional cases. For them HTP can be useful.

I ended up with a good copy of the 24-105 IS, after DxO has finished reducing distortions I have no complaints. All of the RAW converters/sharpening plugins that I have used work fine with the modified 5D, no problems.
Marc

For those interested, I have just uploaded unsharpened ISO 200 crops from each camera in the mini review.

Jack, how does one access these images?

Just click on the link in the first post, but here it is again for ease : http://forum.getdpi.com/forum/showthread.php?t=188&page=3

FWIW, ISO 1600 crops now posted: http://forum.getdpi.com/forum/showthread.php?t=188&page=6

Paprazzi ofter buy the biggest and best cameras, yet quite a few of them use them as point and shoots as they don't care about image quality and actually know very little about photography, bar that selling a picture of a celebrity looking crap can easily pay for such gear.

Jack,
At ISO 1600 I see no practical difference at all between the 1Ds2 and 1Ds3, even at 200% enlargement.

The only thing apparent is that the 5D image is slightly softer. But you might expect it to be because it's been uprezzed. After a modest amount of sharpening, 100% at 0.6 pixel radius, the 5D image also looks on a par with the other two for all practical purposes.

A 200% enlargement on a computer monitor represents a really huge print.

Interesting images. Based on them, I would rather add a Hasselblad to my armoury than a 1DsIII as there appears to be so little difference compared to the 5D. And for that much extra, I'd expect a big difference.
And I know a Hassie doesn't do 1600 before anyone points that out.

JJ, fine - paparazzi are not "complete novices" - they may be something else, but not novices! BTW, did anyone see the zoo-full of them outside the Chicago courthouse where Conrad Black was sentenced today?

To put your comment and Ray's in perspective, we need to recognize that the 5D produces excellent image quality, and it has 12 MP on a FF sensor. Along comes the 1Ds3 with 21 MP on the same size sensor. The fact this was accomplished with at least equivalent image quality is an accomplishment in its own right. In technical terms it looks like a free lunch on quality for 38% more resolution. So I see this as win-win and substantial technical progress, provided these results and conclusions are reliable and representative.

Some one some where (either here or where Jack posted the images) complained about no apparent improvement of DR. Canon never said the DR would increase. They said the move from 12 to 14 bits would improve tonal gradations. That would obviously be a subtle change to pinpoint in a print without careful comparison of images having delicate scales of gradations in them.