hints of maze artifacting?

IR is not accessible right now from where I am. Looks like the site is down.

It was a low ISO shot (100 or 200) of the standard still life scene with bottles, mug, tissue and pencils. I ascribed the artifact to ACR being alpha for the 7D.

I'll process it once again as soon as I see the site up again.

GaborSch said:

Can you post a crop showing the location and the file name?

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

Click to expand...

n/t

GaborSch said:

Note the different averages on the the two green channels of the CFA (caution with "channel"): 2106 vs. 2152. These are normally very close to each other, within a few levels in 14bit.

Click to expand...

Are those levels after subtracting the offset 2048? If so, it suggests that the banding issue is not one of differing offsets for the readout channels but different gains. That means that the banding issue is not just an issue of deep shadows but rather will arise in all exposure zones.

--
emil
--



http://theory.uchicago.edu/~ejm/pix/20d/

bronxbombers said:

took a few lower ISO shots and things have just looked weird, been trying to ignore it and figuring it is just the noise per pixel with the 7D high density but certain things have jsut been looking and acting off, doesn't this look kinda of wormy and mazy? I know ACR is beta for the 7D, but some of these 'noise grains' are like long straight lines that don't seem gaussian at all, zoomed in crop for easier viewing:



i wonder if the calibration on the sensor just isn't off a little with the channels?

Click to expand...

I have never heard or read anywhere where the sensor can be calibrated with the channels. Please explain what this procedure is or where any info on this on thi can be found.

Dietmar

bronxbombers said:

granted ACR can be a bit wormy like that, so maybe it is just that, but it seems to be a bit more than i'm used to seeing.

Click to expand...

Dietmar Fichter said:

I have never heard or read anywhere where the sensor can be calibrated with the channels. Please explain what this procedure is or where any info on this on thi can be found.

Click to expand...

Periodic scalar banding like this seems to be the same at all ISOs, and always with the same pattern and position in the sensor; IOW, it is fixed, and constant. Therefore, before the RAW is written, the 14-bit data could be multiplied by a factor on either odd or even lines to completely rectify the problem. Canon does not seem to care, or try, and with their own converters fixing the problem after the fact, it may be even harder to get them to care.

--
John

ejmartin said:

GaborSch said:

Note the different averages on the the two green channels of the CFA (caution with "channel"): 2106 vs. 2152. These are normally very close to each other, within a few levels in 14bit.

Click to expand...

Are those levels after subtracting the offset 2048? If so, it suggests that the banding issue is not one of differing offsets for the readout channels but different gains.

Click to expand...

That's what I have been referring to as "scalar banding".

Of course, if the gain is applied to the signal with offset, then it will cause a small discrepancy in blackpoint as well.

Ah, if both of these factors were correct, out-of-camera, for all lines! What better images we would have, in all converters.

ejmartin said:

That means that the banding issue is not just an issue of deep shadows but rather will arise in all exposure zones.

Click to expand...

Subjectively worse in the highlights, with more photon-related SNR.

--
John

John Sheehy said:

...before the RAW is written, the 14-bit data could be multiplied by a factor on either odd or even lines to completely rectify the problem. Canon does not seem to care, or try, and with their own converters fixing the problem after the fact, it may be even harder to get them to care.

Click to expand...

Or alternatively, the scaling factors could be determined by factory calibration, and then written into RAW file metadata for the RAW converters to use.

To what degree does this problem show up in JPEGs? Is there any evidence of the bad cameras dealing with the imbalances 'in camera'?

Here is a crop from file E7DhSLI00100_NR_OFF.CR2

Only loaded in lightroom and sharpened 43 0.7 100

The artifacting is all over the place on this raw file.

I don't upscale so as not to incriminate upscaling.

Thank you, now I learned something new.

Dietmar

John Sheehy said:

Dietmar Fichter said:

I have never heard or read anywhere where the sensor can be calibrated with the channels. Please explain what this procedure is or where any info on this on thi can be found.

Click to expand...

Periodic scalar banding like this seems to be the same at all ISOs, and always with the same pattern and position in the sensor; IOW, it is fixed, and constant. Therefore, before the RAW is written, the 14-bit data could be multiplied by a factor on either odd or even lines to completely rectify the problem. Canon does not seem to care, or try, and with their own converters fixing the problem after the fact, it may be even harder to get them to care.

--
John

Click to expand...

1. they are the original pixel values,

2. those captures are from a 5D2, the black level is about 1024,

3. the averages of the two greens are normally almost identical. For example on a patch where the green intensity is between 2059 and 2252 (both channels together), one average is 2140, the other is 2144. (ISO 100, like the one in my demo above).

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

GaborSch said:

1. they are the original pixel values,

2. those captures are from a 5D2, the black level is about 1024,

3. the averages of the two greens are normally almost identical. For example on a patch where the green intensity is between 2059 and 2252 (both channels together), one average is 2140, the other is 2144. (ISO 100, like the one in my demo above).

Click to expand...

Do you have examples of production 7D images which show this sort of banding outside of near-black? I am curious as to whether the problem is one of different offsets or different gains among the readout channels. If the banding is confined to deep shadows, it's an offset problem; if it's in all tonal ranges, it's a problem of different gains.

--
emil
--



http://theory.uchicago.edu/~ejm/pix/20d/

Well, as far as I've seen it is. As measured as percentage of linear exposure it seems to be pretty linear all the way up through the range. There's an offset at blackpoint, but the interline differences then grow with exposure.

ejmartin said:

Do you have examples of production 7D images which show this sort of banding outside of near-black?

Click to expand...

Only in the very deep shadows (how far does "near-black" go?).

See my http://forums.dpreview.com/forums/read.asp?forum=1019&message=33205368 ACR generated sample. The OP posted the raw files, the rapdshare download is still available.

ejmartin said:

I am curious as to whether the problem is one of different offsets or different gains among the readout channels. If the banding is confined to deep shadows, it's an offset problem; if it's in all tonal ranges, it's a problem of different gains

Click to expand...

IMO it is an offset problem, i.e. it could be solved by an addition, not by multiplication. I base this on the fact, that I can't observe proportionally great differences within the 8-pixel wide "bands" in the higher intensity patches.

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

Yes - I'd describe it as a 'basket-weave' texture effect.

It would be very helpfull if you could cut out and post up another clean sample from the same test target, but from a different camera - a nice Nikon for example (I dare you ).

Seriously now - has this kind of effect ever been observed in other brands, like Nikon?

It would surely make some noise measurements, that the likes of DPR perform, fail miserably.

Of course I presume DPR don't take their test cameras from randomly procurred shop bought boxes, do they?

It's very, very, likely that Canon would take extra care to supply the better behaved specimens, which may not truely reflect the same quality variations and randomness that is to be found in the shops.

Mark H said:

Yes - I'd describe it as a 'basket-weave' texture effect.

It would be very helpfull if you could cut out and post up another clean sample from the same test target, but from a different camera - a nice Nikon for example (I dare you ).

Click to expand...

No need to go as far as Nikon - other canon bodies don't make this kind of artefacting. But again, this has to be ascribed to an alpha 7D support of the current ACR.

Mark H said:

It would surely make some noise measurements, that the likes of DPR perform, fail miserably.

Click to expand...

Assuming that the structure is true and not created by ACR.

JulienA said:

When test-driving LR2.5 I downloaded a 7D sample from there and the ACR artifacting shows up for sure.

Click to expand...

yeah I think the IR samples are worse than mine for sure too

most copies probably don't have the issue to any noticeable degree at all even if you use ACR with any settings you possibly wish.

just like with the 5D2 some had it so badly you couldn't avoid it no matter what you did while most (like probably atleast 4999 out of every 5000, if not more) copies you barely see a hint no matter how you tried to manipulate it in ACR.

Mark H said:

Yes - I'd describe it as a 'basket-weave' texture effect.

It would be very helpfull if you could cut out and post up another clean sample from the same test target, but from a different camera - a nice Nikon for example (I dare you ).

Seriously now - has this kind of effect ever been observed in other brands, like Nikon?

It would surely make some noise measurements, that the likes of DPR perform, fail miserably.

Of course I presume DPR don't take their test cameras from randomly procurred shop bought boxes, do they?

It's very, very, likely that Canon would take extra care to supply the better behaved specimens, which may not truely reflect the same quality variations and randomness that is to be found in the shops.

Click to expand...

look at their 5D2 sample from the same scene, nice and smooth, no basket weaving

I remember the first time someone posted this shot (the 7D one) from IR that it was titled "Amazing Sharpness" or something like that and went on about the 7D revealed such amazing fabric texture. Now on the one hand, it did show that the 7D could grab very fine details, that said much of the initial glance wow that is sharp was really just coming from all the harsh basket weave transitions. And if you compare to the 5D2, you'll see that much of the supposed fabric detail capture was really just basket weave/maze artifacts that maybe seemed like a fabric weave but were not at all. And if you looked more closely you'd notice that the basket weave actually all but obliterated the fine graining in the wood holding up the yarn and it somewhat obscures the fine dimpled textures on the white wall (both show very clearly with their 5D2 sample).

It is a little hard to judge but I think my copy looks maybe a touch worse than half-way between the 5D2 smoothness and the IR 7D. I"ll let Gabor have a look once I get the RAWs uploaded since he has an extensive 7D database to compare it with. I'm curious to see if mine really is worse than an avg 7D or not.

JulienA said:

Mark H said:

Yes - I'd describe it as a 'basket-weave' texture effect.

It would be very helpfull if you could cut out and post up another clean sample from the same test target, but from a different camera - a nice Nikon for example (I dare you ).

Click to expand...

No need to go as far as Nikon - other canon bodies don't make this kind of artefacting. But again, this has to be ascribed to an alpha 7D support of the current ACR.

Member said:

It would surely make some noise measurements, that the likes of DPR perform, fail miserably.

Click to expand...

Assuming that the structure is true and not created by ACR.

Click to expand...

The IR one, the effect is so blatant that I tend to think it is the camera. Especially since the final version of ACR for the 5D2 did nothing to remove this effect from the very few 5D2 copies that had it. It does seem that DPP has some degree of maze-control built-in though (although i wonder if some of the low ISO black dot dandruff is not a left over side-effect??).

Using any degree of chroma NR in ACR makes the maze effect worse, as does lowering the masking in the sharpening part of ACR. Putting masking to 100 and luma noise to 100 and chroma noise to 0 erases all of it (although getting tht dramatic with the settings also makes things look really hideously smoother over).

bronxbombers said:

The IR one, the effect is so blatant that I tend to think it is the camera. Especially since the final version of ACR for the 5D2 did nothing to remove this effect from the very few 5D2 copies that had it. It does seem that DPP has some degree of maze-control built-in though (although i wonder if some of the low ISO black dot dandruff is not a left over side-effect??).

Click to expand...

That would be really worrying if it is the camera. The IR files have a "cooked up", unnatural look in ACR that makes this converter useless if it's not fixable.