Another last test... (Xt vs. Xti)

dciobota · Oct 25, 2006

I have an almost 2 year old 350D and just got my 400D on friday. I did some testing between the two and posted my impressions and results in this thread:

http://forums.dpreview.com/forums/read.asp?forum=1031&message=20562210

As far as I can tell, if you take the metering out of the picture, the sensors expose essentially the same. My XTi does seem to have a tad more noise at longer exposures, but not something I consider deal stopping (about 1/3 stop, the way I measured it).

That's good enough for me.

Daniel

--
----------------------------------------------
http://www.pbase.com/dciobota

John Sheehy · Oct 25, 2006

Ciriaco Garcia said:
The photons falling in both sensors are just
the same, both in wavelenght and quantity, in average, and the 350D
pixels are clearly "more filled" using an identical exposure.

You can't tell directly how the pixels are filled from a histogram of the RAW data. There are two other variables: how many photons equal 1 RAW level, and what percentage of the well capacity is each RAW level, at a given ISO, for any given camera model.

--
John

John Sheehy · Oct 25, 2006

Ciriaco Garcia said:
The XTi should have occupy a 25% more (the additional resolution)
but instead it has taken a 57,3% more space. This "26% shift" in
the file size is an excelent unbiased measure of the additional
noise.

Could be from a difference in image sharpness, too. It could also be that, unlike most other Canon DSLRs, the XT actually has some traditional software noise reduction in its RAW files. The RAW samples I downloaded from this thread certainly look that way to me; the combined noise and detail are softer for the XT. If it were caused by a strong AA filter or softer optics, the noise would still be high-frequency, but the XT noise looks like it is weak in high-frequency content (at the nyquist).

When comparing sharp noise to dull noise, the dull noise may look just as strong at 100% pixel view, but when reduced in size by small printing or downsampling, the sharper noise clears up more, as it fails to resolve as well, or averages (or diffuses) away better.

--
John

John Sheehy · Oct 25, 2006

Glenn W. said:
You know... while there clearly is a differance between these two
perticular cameras... and maybe theres a simmilar differance
between all XT and XTi cameras... I still can't help but think that
the XTi is still taking a VERY nice photo... To some extent I
think all this concern over this 0.4 stops of exposure differance
is really overblown. Now I also know that to the person who's just
upgraded from Xt to Xti, this may seem like a big issue that really
bothers them... So I do see both sides of the issue...

However... You know as a 300D owner... (original Digital Rebel)...
Man I'd love to have ISO 1600 that clean... Heck my ISO 400 is
barely that good....

The 10D/300D was the last Canon DSLR before major innovations in high-ISO noise. The standard deviation of a 10D/300D blackframe at ISO 1600 is 18.0. It's 7.2 for the XTi, 4.7 for the 20D/30D, about 5.4 for the 1DSmkII (IIRC), 4.8 for the 1DmkII, and 4.6 for the 5D. There is a little difference in RAW sensitivity, as we are seeing in this thread, so these numbers need to be scaled a little to reflect the absolute exposure's noise floor at ISO 1600.

--
John

Ciriaco Garcia · Oct 25, 2006

The -b parameter to adjust the brighness in dcraw is "linear"... this means that I shouldn't have used the square the value, but simply 1/1.31 to adjust the exposure -04 EV. And I was anyway wrong because I should have used the square root, I think

In fact, sometimes people have used the -b 4 switch to compensate the raw conversions in some cameras when there is need to adjust 10 to 12 bit.

Using -b 0.7634 the final XTi file size is 3117171 bytes (a 44% more compared to XT, for a 25% more resolution).

Ciriaco Garcia · Oct 25, 2006

Only to comment that the dcraw conversions I have uploaded were not sharpened... only "dcraw -w" and the ppm converted to jpeg.

John Sheehy said:
Ciriaco Garcia said:

The XTi should have occupy a 25% more (the additional resolution)
but instead it has taken a 57,3% more space. This "26% shift" in
the file size is an excelent unbiased measure of the additional
noise.

Click to expand...

Could be from a difference in image sharpness, too. It could also
be that, unlike most other Canon DSLRs, the XT actually has some
traditional software noise reduction in its RAW files. The RAW
samples I downloaded from this thread certainly look that way to
me; the combined noise and detail are softer for the XT. If it
were caused by a strong AA filter or softer optics, the noise would
still be high-frequency, but the XT noise looks like it is weak in
high-frequency content (at the nyquist).

When comparing sharp noise to dull noise, the dull noise may look
just as strong at 100% pixel view, but when reduced in size by
small printing or downsampling, the sharper noise clears up more,
as it fails to resolve as well, or averages (or diffuses) away
better.

--
John

Ciriaco Garcia · Oct 25, 2006

The ISO 1600 conversions by dcraw I have uploaded, have exactly the same brightness in both cameras, despite the XTi being shifted 0.4 EV to the left. This is because dcraw converter (as many others may do) normalized the underexposed data to use the full range (0-255 in jpeg) and finally both pictures seemed equally exposed. But this was not true in the raw data, and the XTi has finally reached a lot of more noise as a consequence.

If the picture had been more contrasty, and some picture areas had reached the maximum exposure in both cameras (the XT more clipped) then dcraw had likely returned a darker XTi image trying to preserve the existing highlights in both (dcraw has also the -H switch to alter the clipping mode).

The jpeg and the raw conversors are very capable in the task of fooling us. Even looking at the raw data may be subject to different interpretations.

dciobota said:
I have an almost 2 year old 350D and just got my 400D on friday. I
did some testing between the two and posted my impressions and
results in this thread:

http://forums.dpreview.com/forums/read.asp?forum=1031&message=20562210

As far as I can tell, if you take the metering out of the picture,
the sensors expose essentially the same. My XTi does seem to have
a tad more noise at longer exposures, but not something I consider
deal stopping (about 1/3 stop, the way I measured it).

That's good enough for me.

Daniel

--
----------------------------------------------
http://www.pbase.com/dciobota
----------------------------------------------

John Sheehy · Oct 25, 2006

Ciriaco Garcia said:
Only to comment that the dcraw conversions I have uploaded were not
sharpened... only "dcraw -w" and the ppm converted to jpeg.

I was refering to the RAW data itself, actually, from the CR2s.

--
John

Ciriaco Garcia · Oct 25, 2006

Let's imagine that Canon for simplicity hasn't modified the parameters you say in both cameras, but only has placed a new improved sensor as they claim (reducing the gap between pixels) with a few jpeg engine changes. After seen how little they are worried about different calibration among cameras, this speculation is not totally out of sense.

Just for fun, playing with the Excel data, I have sum all the pixel values in the ISO 1600 picture. Well, both pictures aren't exactly equal (hope that statistical science makes them comparable) and these are the results:

XT: 3889382193
XTi: 4126993817
XT average pixel value: 483,82
XTi average pixel value: 406,06

This would mean that the new sensor, despite having the same surface, has a total 6% greater capacity to collect light, but the capacity per pixel has decreased a 16% (only 0,25 stops). This is slightly better compared with the previous 0,4 stops conclusion by looking at the histogram limits.

Yes, a total speculation... they can calibrate the hardware as they want.

John Sheehy said:
Ciriaco Garcia said:

The photons falling in both sensors are just
the same, both in wavelenght and quantity, in average, and the 350D
pixels are clearly "more filled" using an identical exposure.

Click to expand...

You can't tell directly how the pixels are filled from a histogram
of the RAW data. There are two other variables: how many photons
equal 1 RAW level, and what percentage of the well capacity is each
RAW level, at a given ISO, for any given camera model.

--
John

Ciriaco Garcia · Oct 25, 2006

no text

John Sheehy · Oct 25, 2006

Ciriaco Garcia said:
Let's imagine that Canon for simplicity hasn't modified the
parameters you say in both cameras, but only has placed a new
improved sensor as they claim (reducing the gap between pixels)
with a few jpeg engine changes. After seen how little they are
worried about different calibration among cameras, this speculation
is not totally out of sense.

Just for fun, playing with the Excel data, I have sum all the pixel
values in the ISO 1600 picture. Well, both pictures aren't exactly
equal (hope that statistical science makes them comparable) and
these are the results:

XT: 3889382193
XTi: 4126993817
XT average pixel value: 483,82
XTi average pixel value: 406,06

These are real RAW levels? Is blackpoint subtracted? If the answers are yes and no, then you need to subtract 256 from both values, as 256 is black in both cameras. Then, the difference is actually 1.51x, about 2/3 stop.

--
John

austinphotog · Oct 25, 2006

Thanks for the thread.

Let me paraphrase to see if I understand. There is not much difference in IQ between the XT and XTI. At ISO 100-400 the noise is around the same, while at 1600 the XTI is slightly noisier. The XTI appears even noisier since its exposure tends to be less than the XT. So XTI will have slightly better IQ at lower ISOs but possibly worse at 1600. Either way the difference is small.

From those with XTIs would you buy it again. Are the benefits of the focus system and picture styles worth buying it over a lower cost XT?

Kenstrain said:
Thanks again Glenn for hosting these.

To save people reading the whole thread.

These are two examples of many conversions from the raw files
provided by John and hosted by Glenn.

The two images are from the raw files for 1600 ISO. They are
processed using Capture 1 v3.7.6b (beta). This is my usual raw
converter, although not as transparent to use as dcraw or IRIS.
See in this thread and elsewhere for results using dcraw.

I found small adjustments to white balance (mostly in the 50 to
100K range) were needed to match the greys. NR was set to minimum,
sharpening to a minimal amount (there may of course be a little of
both). Curve was "film standard" exposure was compensated by 1.20
stops and 1.65 stops - XT, XTi, respectively in case you did not
guess. (Possibly my earlier report of 0.55 stops difference was a
bit on the high side.)

Processing was meant to bring out the noise rather than hide it.
I also made some with a little more NR and the XTi seemed to retain
the small advantage.

Destination was sRGB for these two images only, compressed only
after resizing. I resised the XT image using a 6x6 Lanczos
algorithm in Picture Windows Pro v 3.5. I did this both ways round
(i.e. also shrinking the XTi image to match the XT one) and the
differences seem to be preserved, so I only provide one set here.

Note that the image sharpness is not exactly the same in these two
images, whether due to the camera or the lens or my processing. is
hard to tell.

I found it interesting to exaggerate noise using edge detection,
but due to the large file sizes will leave that up to the
enthusiast to try.

The differences seemed progressively smaller at 800 and 400 ISO
(nearly indistinguishable in the latter case).

I considered offering crops or lower quality jpegs, but several
crops would be needed to avoid misleading - look at various parts
of the images.

looking at the whole set of images I had, these two seemed the
fairest comparison overall. It is rather easy to change the
processing and make some aspect look a litte better or worse. I
recommend judging by viewing at 1:2 or 1:3 or printing.

Feel free to ask if I omitted something important.

Ken

Ciriaco Garcia · Oct 25, 2006

Yes, you have reason. Dcraw also uses the 258 black point for the 400D, at least by the moment.

By the way, do you know why several Canon cameras uses a different black point?. In particular the 20/30D compared to 350/400D.

I look some time ago at dcraw's code at it seemed to me that it uses some of the "hidden" pixels (e.g. the 400D image is 3948 pixels width, and dcraw only uses 3906). I think that it uses these 32 pixels to compute the black point (!), which sounds me very strange, but I should take a more careful look to be completely sure (maybe the author uses the 'black' variable in the code for more purposes).

John Sheehy said:
Ciriaco Garcia said:

Let's imagine that Canon for simplicity hasn't modified the
parameters you say in both cameras, but only has placed a new
improved sensor as they claim (reducing the gap between pixels)
with a few jpeg engine changes. After seen how little they are
worried about different calibration among cameras, this speculation
is not totally out of sense.

Just for fun, playing with the Excel data, I have sum all the pixel
values in the ISO 1600 picture. Well, both pictures aren't exactly
equal (hope that statistical science makes them comparable) and
these are the results:

XT: 3889382193
XTi: 4126993817
XT average pixel value: 483,82
XTi average pixel value: 406,06

Click to expand...

These are real RAW levels? Is blackpoint subtracted? If the
answers are yes and no, then you need to subtract 256 from both
values, as 256 is black in both cameras. Then, the difference is
actually 1.51x, about 2/3 stop.

--
John

John Sheehy · Oct 25, 2006

austinphotog said:
Thanks for the thread.

austinphotog said:
Let me paraphrase to see if I understand. There is not much
difference in IQ between the XT and XTI. At ISO 100-400 the noise
is around the same, while at 1600 the XTI is slightly noisier. The
XTI appears even noisier since its exposure tends to be less than
the XT. So XTI will have slightly better IQ at lower ISOs but
possibly worse at 1600. Either way the difference is small.

IMO, there is a big difference in comparing noise at the highest ISO, and at the lowest ones. The highest ISO is used not only for good exposure, but also in extreme under-exposure, so for ISO 1600, the real-world S/N ratio is very critical. For the lowest ISOs, one usually has ample light and actual sensitivity is not really important, just getting a good relative exposure for the ISO, and file format (JPEG or RAW). The XTi's ISO 100 may meter for 120, and actually be about 80, but if you expose for 80 (or expose "200" for 160) you have the least noise of any APS-sized Canon DSLR, because these ISOs have the least readout noise, relative to full RAW saturation, and therefore the most DR.

--
John

Green_Thark · Oct 25, 2006

Glenn W. said:
You know... while there clearly is a differance between these two
perticular cameras... and maybe theres a simmilar differance
between all XT and XTi cameras... I still can't help but think that
the XTi is still taking a VERY nice photo... To some extent I
think all this concern over this 0.4 stops of exposure differance
is really overblown. Now I also know that to the person who's just
upgraded from Xt to Xti, this may seem like a big issue that really
bothers them... So I do see both sides of the issue...

However... You know as a 300D owner... (original Digital Rebel)...
Man I'd love to have ISO 1600 that clean... Heck my ISO 400 is
barely that good....

Good point, but Im sure John August would like to know if this is normal behaviour or if his camera is under sensitive for some reason. I'm a lot more happy than John because using the same software at the same settings I am getting a smaller difference between my xt & xti so is it just normal camera variation maybe John has the lower end sensitive XTI and higher end sensitive XT and me the opposite or is there something wrong with his XTI, all in all a tricky question to answer.

John Sheehy · Oct 25, 2006

Ciriaco Garcia said:
Yes, you have reason. Dcraw also uses the 258 black point for the
400D, at least by the moment.

By the way, do you know why several Canon cameras uses a different
black point?. In particular the 20/30D compared to 350/400D.

Why they're not zero, or why they are different?

The fact that they're not zero has a beneficial effect for any software that tries to remove fixed pattern noises or shadow banding (not that many even try), and are better for stacking and binning as clipping the black to zero would make the average level of a black area non-black, and therefore the extreme shadows are slightly noisier and non-linear (most other brands clip at black, AFAIK; Nikon does).

I don't know why Canon does it, though. It may just be an accidental benefit. They don't use the info to any useful effect in their software, AFAICT. I have no idea why different cameras are different, but some are different at different ISOs, like the 10D.

Ciriaco Garcia said:
I look some time ago at dcraw's code at it seemed to me that it
uses some of the "hidden" pixels (e.g. the 400D image is 3948
pixels width, and dcraw only uses 3906). I think that it uses these
32 pixels to compute the black point (!), which sounds me very
strange, but I should take a more careful look to be completely
sure (maybe the author uses the 'black' variable in the code for
more purposes).

Yes, that's the most accurate way of getting the blackpoint, from those pixels. ACR uses them to remove line-banding from some cameras, which removes strong banding effectively, but leaves light banding in the image.

If Canon had the wisdom to make these dark areas just a little wider, and on all 4 sides, instead of two, the banding could be removed more completely, by software that promoted the RAW data to at least a couple more bits.

--
John

Kenstrain · Oct 25, 2006

austinphotog said:
There is not much
difference in IQ between the XT and XTI. At ISO 100-400 the noise
is around the same, while at 1600 the XTI is slightly noisier. The
XTI appears even noisier since its exposure tends to be less than
the XT.

From my very limited experience, I would say that the difference is tiny at 1600, including the "exposure issue" - from my point of view it makes little sense to say what the noise is on images that are not adjusted to look similar.

It is a bigger jump from a 300D/Rebel, but still not huge.

austinphotog said:
From those with XTIs would you buy it again.

Yes, it is a great little camera....

... I don't have an XT - that is why I was processing John's files.

Ken

Sabz · Oct 25, 2006

Given the number of under-exposure treads in this forum, it would be a service to the forum members if you could put this information on a web page so that that we don’t end up hashing the same issue over and over again but if anything maybe as we learn me can add to it.

As far as “variation among XTi samples” are concerned, I have to disagree. Due to series of "unfortunate accidents" ended up testing 3 samples from two different merchants. The only variation that I found was the number of stuck pixels that varied more than I expected. Otherwise the underexposure issue was quite the same in all samples (I wish that I had kept my test photographs).

As a person that have moved from XT to XTi, I have a love and hate relationship with this camera. Exposed correctly, one can get quite stunning photographs (look to me better than XT photographs). However, I am discouraged when due to exposure inconsistency, I have to adjust the exposure by more than +1 ev and see higher than expected noise, even in my 200 and 400 ISO photographs. The focus speed and accuracy is without a doubt much better.

All and all this is not a camera for point and shooters. Not yet at least until/if this issue is resolved via firmware upgrade. Given that the lower sensitivity will always be there.

dciobota · Oct 25, 2006

Well, I suspect raw converters don't all act alike, and I was suspicious of acr3.1 vs acr3.6 conversion between the two models.

But... if you notice, the scene I shot, and on purpose, has a much higher dynamic range than either camera can record. Look at the window to the right, it is totally blown out in the pic. The shadow areas on the left of the cabinet are also cut to black. The histogram shows that clipping very clearly on both ends of the scale. So, I don't think in either case the raw converter compensated.

Also, the settings I used were to specifically prevent the raw converter to mess with the original data, as explained. I specifically used a linear curve, so no Canon specific curve could be applied (ie like picture styles and such), no enhancements, brightness at 50%. It's as "flat" as you're going to get out of the camera without going through some esoteric raw interpreter, which as you say, can debayer and interpolate the data its own way.

Bottom line. Take a pic through the camera, use as identical a set of settings and environments to convert your data, and the final result is pretty much the same. That's where it counts anyways, and I look at the end result to make an objective comparison with my equipment.

Of course, ymmv. ;-)

Daniel

Ciriaco Garcia said:
If the picture had been more contrasty, and some picture areas had
reached the maximum exposure in both cameras (the XT more clipped)
then dcraw had likely returned a darker XTi image trying to
preserve the existing highlights in both (dcraw has also the -H
switch to alter the clipping mode).

The jpeg and the raw conversors are very capable in the task of
fooling us. Even looking at the raw data may be subject to
different interpretations.

dciobota said:

I have an almost 2 year old 350D and just got my 400D on friday. I
did some testing between the two and posted my impressions and
results in this thread:

http://forums.dpreview.com/forums/read.asp?forum=1031&message=20562210

As far as I can tell, if you take the metering out of the picture,
the sensors expose essentially the same. My XTi does seem to have
a tad more noise at longer exposures, but not something I consider
deal stopping (about 1/3 stop, the way I measured it).

That's good enough for me.

Daniel

--
----------------------------------------------
http://www.pbase.com/dciobota
----------------------------------------------

Click to expand...

--
----------------------------------------------
http://www.pbase.com/dciobota

Rick rock · Nov 4, 2006

Hi, John

Thanks for your answer to my badly written post on the Open Talk Forum, I changed my reply to this trend since the "open talk" doesn't seem to be so open after all, regarding some posts I have seen.

I´ve read something about raw noise processing in the xt being somewhat higher at raw level. Since I always assumed raw data was free from noise processing, if this be true, how does it compare to the 20D raw processing?

Can we make a direct comparison between the 20D and the XT in order to reveal somewhat different raw capture choices from Canon, considering the same basic sensor is used on both cameras?
Kind Regards

Rick rock said:
In my opinion, this situation really needs an explanation, because
it may look like somebody is not very happy about exposing this
data to everyone.

The XTi, with the same ISO and Tv and Av settings, has RAW data 1/2 stop less exposed (72% the value) of my 20D.

With equal exposure like this, the signal-to-noise-floor ratio is 2.14x as high in the 20D at ISO 1600, and 1.1x as high at ISO 100. The 400D has higher DR at ISOs 100 and 200. This is all at the pixel level, and the finer noise grain of the 400D gives it a boost in all these categories.

In my estimation, the 20D meters for about 1.19x the stated ISO, and actually has a standard 3.5 stops of headroom, making its RAW sensitivity what it meters at. The XTi meters for 1.19x 85% the stated ISO as well, but its actual RAW sensitivity is about 86% of the stated ISO.

In other people's similar tests with XT vs XTi, the XT seems to be about 2/3 stop higher RAW sensitivity, and the XT meters liberally for higher RAW exposures than the 20D and XTi.

Another last test... (Xt vs. Xti)

Leading Member

Forum Pro

Forum Pro

Forum Pro

Forum Enthusiast

Forum Enthusiast

Forum Enthusiast

Forum Pro

Forum Enthusiast

Forum Enthusiast

Forum Pro

Leading Member

Forum Enthusiast

Forum Pro

Leading Member

Forum Pro

Forum Enthusiast

Member

Leading Member

Member

Similar threads

Keyboard shortcuts