Canon 7D vs 5D2 noise in DxO Mark

[Guillermo Luijk]

I was just comparing the recently uploaded analysis of the Canon 7D in DxO Mark.

Noise in the deep shadows (dynamic range) in the 7D seems to match that of the 5D2 at low ISOs (100-200), being better in the 5D2 by 1 stop or more at high ISOs:




If that were true the 7D would be performing fantastic. BUT if we take the 18% SNR measurements (noise in the correctly exposed areas), I find a too big advantage for the 5D2 compared to the previous measurement:



Both graphs considered in 'Print' mode (i.e. taking into account pixel count).

Gabor, did you already analyze colourcheker patches on the 7D?

Salu2

[thierrylegros396]

So, it's absolutely normal !

That also means ETTR is more important with the D7 than with the 5D2.

Curious to see a comparison between the G10 and G11 to see if we have same difference.


Have a Nice Day.

Thierry

[Panopeeper]

Guillermo,

the comparison with 18% is IMO a totally worthless excercise. Who cares? That level corresponds to about -2.5 EV from clipping - I don't see any noise even with ISO 1600 at that level. Is is important, if the SNR is 50 or 100? I measured the noise at that level with ISO 100; the 5D2 shows 1.2% (SNR=83), the 7D shows about 2.2% (SNR=45). Who cares?

On the other hand, the comparison with SNR=1 is nonsensical for the opposite reason. 100% noise looks horrendeous, it is accepted only with heavy noise reduction, and only when there is no other way. I don't even measure over 75% (as the matter of fact, I doubt they measured that properly). I found about 0.35 EV advantage of the 5D2 with ISO 100, measured at the intensity around nine stops from clipping. The difference is apparently around 0.7 EV with ISO 1600; I don't have suitable raw files from the 7D.

[ErikKaffehr]

Guillermo,

I compared 7D with 20D and 40 in DxO-mark and was surprised to see how small the difference was. Have you tried that?

Best regards
Erik

I was just comparing the recently uploaded analysis of the Canon 7D in DxO Mark.

Noise in the deep shadows (dynamic range) in the 7D seems to match that of the 5D2 at low ISOs (100-200), being better in the 5D2 by 1 stop or more at high ISOs:




If that were true the 7D would be performing fantastic. BUT if we take the 18% SNR measurements (noise in the correctly exposed areas), I find a too big advantage for the 5D2 compared to the previous measurement:



Both graphs considered in 'Print' mode (i.e. taking into account pixel count).

Gabor, did you already analyze colourcheker patches on the 7D?

Salu2

[Guillermo Luijk]

Guillermo,

the comparison with 18% is IMO a totally worthless excercise. Who cares? That level corresponds to about -2.5 EV from clipping - I don't see any noise even with ISO 1600 at that level. Is is important, if the SNR is 50 or 100? I measured the noise at that level with ISO 100; the 5D2 shows 1.2% (SNR=83), the 7D shows about 2.2% (SNR=45). Who cares?

On the other hand, the comparison with SNR=1 is nonsensical for the opposite reason. 100% noise looks horrendeous, it is accepted only with heavy noise reduction, and only when there is no other way. I don't even measure over 75% (as the matter of fact, I doubt they measured that properly). I found about 0.35 EV advantage of the 5D2 with ISO 100, measured at the intensity around nine stops from clipping. The difference is apparently around 0.7 EV with ISO 1600; I don't have suitable raw files from the 7D.

For properly exposed low and medium DR scenes, -2,5EV from clipping is the zone where most of the RAW information will fall so it's the proper benchmark to know how the camera will behave (it takes into account photon noise vs full well capacity). I agree is not a critical area at base ISO though.

However I agree with you it's worthless in terms of measuring noise in really hard conditions (HDR situations and/or very low lighting). I simply was surprised at the big difference gap for that particular camera in 18% area vs deep shadows.

Now I had more time and went to see the SNR curves which should be the source of the story, and was disappointed to see they are plotting different curves for non-electronic ISOs (fake ISOs) so I won't take them too seriously. For example curves for ISO6400 and above are different to the ISO3200 curve in the Canon 5D2 plot:



I will have some colorcheker shots for the 7D soon, will let you know if you are interested.

Regards

[Guillermo Luijk]

I compared 7D with 20D and 40 in DxO-mark and was surprised to see how small the difference was. Have you tried that?

I have tried it and results are expected to me. Do you find something strange there? the 7D shows a fairly good improvement over the 40D.

[Panopeeper]

For properly exposed low and medium DR scenes, -2,5EV from clipping is the zone where most of the RAW information will fall so it's the proper benchmark to know how the camera will behave (it takes into account photon noise vs full well capacity). I agree is not a critical area at base ISO though

It is not critical for anything relevant. For example the 7D's noise at -2.5 EV is 7.2% with ISO 3200 and 10% with ISO 6400.

Now I had more time and went to see the SNR curves which should be the source of the story, and was disappointed to see they are plotting different curves for non-electronic ISOs (fake ISOs) so I won't take them too seriously

Like DPReview, which often find higher dynamic range with increasing ISOs, or different DR with different fake ISOs. For example the 40D has the same DR, 8.9 EV with ISO 400, 800, 1600, and then 7.4 EV (1.5 EV lower!) with the fake 3200.

ADDED

I took a second look at those graphs. The x axis is the intensity, obviously measured relatively to the clipping, i.e. 100% is the clipping value. One stop higher fake ISO works with half of the exposure, i.e. half of the light gets captured, and it doubles the pixel values. Thus the same % with ISO 12600 represents half as much captured light as with ISO 6400, and that causes higher noise/lower DR.

I will have some colorcheker shots for the 7D soon, will let you know if you are interested

I am interested, if they are sufficiently underexposed. Thanks.

[Guillermo Luijk]

I took a second look at those graphs. The x axis is the intensity, obviously measured relatively to the clipping, i.e. 100% is the clipping value. One stop higher fake ISO works with half of the exposure, i.e. half of the light gets captured, and it doubles the pixel values. Thus the same % with ISO 12600 represents half as much captured light as with ISO 6400, and that causes higher noise/lower DR.

That could be an explanation, their X-axis is amount of light (real exposure) instead of RAW exposure. But then it is not clear why all the curves from ISO3200 are not replicas of the ISO3200 curve uniformly shifted 1 stop, because they should.

On the other hand I am not sure if 100% is the clipping value (which would be the right way) or the max in the RAW scale, which for many cameras is higher than the clipping level.

Expecting the RAW files from the 7D I received a pair of colourchecker shots from the 5D2 and plotted the SNR curves (red is ISO100, blue is ISO1600):



Measured DR (SNR=12 dB criteria):
- ISO100: 9,0 stops
- ISO1600: 7,0 stops

Extrapolated DR (SNR=0 dB criteria, used by DxO Mark):
- ISO100: 11,something stops (DxO Mark gives 11,16 stops)
- ISO1600: nearly 10 stops (DxO Mark gives 10,05 stops)

[feppe]

Guillermo,

I compared 7D with 20D and 40 in DxO-mark and was surprised to see how small the difference was. Have you tried that?

Best regards
Erik

I did the same comparison between 450D and 7D. While dynamic range is better with 7D, SNR, Tonal range and Color sensitivity are better with the 450D in screen mode. If I go to print mode, the differences all but disappear. I'm not entirely sure how to read that, though.

Does it imply that although the 7D has a higher dynamic range than the 450D, the extra stop or so is of poor quality? Does it also imply that the higher pixel count on the 7D compensates for the poorer per-pixel quality?

I've been considering getting the 7D for the UI improvements and potential IQ upgrade, but it seems that the latter is only marginal, driven by higher pixel count.

[ErikKaffehr]

Hi,

Yes a small advantage for the 7D over the 20D regarding DR but a pretty small difference regarding the other curves.

Best regards
Erik

I have tried it and results are expected to me. Do you find something strange there? the 7D shows a fairly good improvement over the 40D.