[TonyW]

Sure. I suppose I should spend more time looking at actual images and rely much less on test charts. At the end of the day, with images, what you get is what you see, or what you see is what you get. KP images look fantastic.

Absolutely agree look at test charts by all means but also very important if you can get hold of real images raw and JPEG. Base ISO always the best starting point and if you wish to look at different ISO then both raw and JPEG may be of interest to see what the manufacturers do with the data.

But then you need to have trust in the testing methodology and consistency. It is also my firmly held view that the raw data needs to be treated properly to get the best out of the images including capture sharpening, noise reduction etc. In comparison you may wish to also look at equalising sizes both upsampling and downsampling

In addition make sure that raw defaults are set to zero on import of raw files for your comparisons

FWIW and being curious I had a look at the K1 MKII and the Sony at base and at ISO 1600. At Base ISO100 noise characteristics difference seem negligible. At ISO1600 Sony more noise but that can be corrected in raw if seen as a real problem - small amounts of chroma and luminence adjustments will correct without unduly affecting resolution. First image ISO100 second ISO1600.

Attached Images

   

[biz-engineer]

The way I understand dual gain is that there is a gain switch before the A/D conversion, so that pixel signals get a boost before conversion at higher ISO. If this is so, the advantage of dual gain should be very small at 400 ISO, and more significant at very high ISOs as read noise becomes more prominent. What I'm seeing in curves in a strong bump after the gain is supposedly switched, which is probably not only coming from gain switch in the hardware but also some kind of data processing (like noise reduction) in camera firmware. Anyway, according to the PDR curve, the bump of PDR for the A7RIV is like going back to 2 stops lower ISO on the curve after the dual gain kicks in, a 2 stop improvement in dynamic range which basically is impossible just be reducing the read noise of a 14bits A/D converter.

[UncleVanya]

The way I understand dual gain is that there is a gain switch before the A/D conversion, so that pixel signals get a boost before conversion at higher ISO. If this is so, the advantage of dual gain should be very small at 400 ISO, and more significant at very high ISOs as read noise becomes more prominent. What I'm seeing in curves in a strong bump after the gain is supposedly switched, which is probably not only coming from gain switch in the hardware but also some kind of data processing (like noise reduction) in camera firmware. Anyway, according to the PDR curve, the bump of PDR for the A7RIV is like going back to 2 stops lower ISO on the curve after the dual gain kicks in, a 2 stop improvement in dynamic range which basically is impossible without some sort of noise processing.

That kind of change also happens with the KP. Clearly it is noise processing related but that doesn’t make it invalid.

[stevebrot]

Test charts not agreeing with what I see

One may assume that either what one sees is wrong or the charts do no say what we think they say or the charts are wrong.

I don't use either site because I don't believe the charts say what they claim, at least not in terms of the language/terms used.


Steve

[JensE]

That sadly is marketing b/s.The Sony dual gain sensors deliver no more detail, they lag behind actually at given lighting situations.The a7r3 is actually the worst performer with regards to ISO invariance of those three cameras. The ten year old K-5 (within its apsc limits) outshines an a7r3 on ISO invariance until ISO 1600, same as a Nikon D7000.Looking at the actual images easily reveals this.

Sorry, I didn't express that clearly. I was discussing PDR measurements and why I think that dual gain makes sense for the A7RIII or IV sensors. And, again, for a sensor with higher noise than the excellent K-1 (and K-5) at base ISO, why strict ISO invariance is not desirable. If you start from a good high-ISO performance and extrapolate that to low ISO in the sense that "the low noise level is maintained even at base ISO", it surely is. But for sensors with mediocre low-ISO dynamic range, such as older Canon sensors, you do want to reduce the (absolute) noise level as you increase ISO, so that your dynamic range can become competitive at high ISO - one reason why Canon cameras were competitive e.g. for sports photographers.


What I meant to say is that in the PDR over nominal ISO, the A7RIII measures better only due to the switchable gain. Otherwise either low-ISO or high-ISO would suffer. I don't have access to any A7Rxx, and can only observe published shots. Comparing the low-light DPR shots visually at higher ISO, I would say that K-1 and A7RIII are head-to-head there at the same nominal ISO in terms of shadow detail: Some areas look better for one, other areas for the other. Combining that with the observation that ISO is overstated by 1/2 EV (according to DXO), it confirms a stated advantage of the K-1 over the A7RIII. Applying the shift in the PDR curve would make them (only) identical: Again, a testament to my statements above how the measured PDR isn't sufficient to describe the visual quality with respect to noise in detail. PDR is a necessary, but not sufficient criterion for good visual appearance. Some very gentle noise reduction may indeed be at play here.

[UncleVanya]

I also want to understand if comparisons that are referenced in the thread are being done by pixel peeping without normalizing the pixels. A 36mp k-1 should not be directly compared at pixel noise level with a 60mp A7rIV. The pixel counts should be normalized to either upsample the k-1 or downsample the A7rIV - otherwise the difference in resolution could make tiny differences look larger than they are. For reference read this article: Signal to Noise Ratio: Why some camera comparisons are wrong ? Thomas Fitzgerald Photography

[TonyW]

I also want to understand if comparisons that are referenced in the thread are being done by pixel peeping without normalizing the pixels. A 36mp k-1 should not be directly compared at pixel noise level with a 60mp A7rIV. The pixel counts should be normalized to either upsample the k-1 or downsample the A7rIV - otherwise the difference in resolution could make tiny differences look larger than they are. For reference read this article: Signal to Noise Ratio: Why some camera comparisons are wrong ? Thomas Fitzgerald Photography

Exactly. Hence the comments made earlier in this thread
“In comparison you may wish to also look at equalising sizes both upsampling and downsampling”

[UncleVanya]

Exactly. Hence the comments made earlier in this thread
“In comparison you may wish to also look at equalising sizes both upsampling and downsampling”

This is seen with the k-3 vs k-5 - many people were appalled at first then realized there really wasn’t a huge difference, the k-3 noise was lower than they originally thought by simple pixel level comparisons. I think the k-5 still wins in the noise department but it’s a lot closer than originally thought.

[JensE]

The pixel counts should be normalized to either upsample the k-1 or downsample the A7rIV - otherwise the difference in resolution could make tiny differences look larger than they are.

As far as the DPR comparison shots go, the 'Comp' and 'Print' views do exactly that (Comp just downsamples one, Print both). The PDR definition referenced above is also resolution independent.

[UncleVanya]

As far as the DPR comparison shots go, the 'Comp' and 'Print' views do exactly that (Comp just downsamples one, Print both). The PDR definition referenced above is also resolution independent.

I knew they used measurements that took that into account. I’m not sure the ‘downloaded a raw file and compared to my xxx’ comments above were done in that way.

[biz-engineer]

I also want to understand if comparisons that are referenced in the thread are being done by pixel peeping without normalizing the pixels. A 36mp k-1 should not be directly compared at pixel noise level with a 60mp A7rIV. The pixel counts should be normalized to either upsample the k-1 or downsample the A7rIV - otherwise the difference in resolution could make tiny differences look larger than they are.

Yes, comparisons done at the same size. Pentax K-1 slightly less noisy. I don't see how dual gain on the Sony is an advantage , maybe it help bring back down the extra noise generated by faster readout (faster readout is necessary for video , but general more noise from sensor silicon substrate).

---------- Post added 24-02-21 at 07:44 ----------

One may assume that either what one sees is wrong or the charts do no say what we think they say or the charts are wrong.I don't use either site because I don't believe the charts say what they claim, at least not in terms of the language/terms used.

wise statement.

---------- Post added 24-02-21 at 07:54 ----------

And, again, for a sensor with higher noise than the excellent K-1 (and K-5) at base ISO, why strict ISO invariance is not desirable.

Whether ISO invariance is better or not can't be answered without knowing the noise figure of each stage. In radio received, every added amplification stage in the analog domain adds noise and the first amplification stage in the signal chain is the one that contribute most to total noise, that's why the first amp is usually an amplifier optimized for noise (so called LNA). However, digital amplification doesn't add any noise. So that pause the question of image sensors. In principle, the less amp in the signal chain the lower the noise , except if one element is noisy than it may benefit to add a lower noise amp stage in front to mitigate the effect of the noisy stage. In other words, if A/D conversion must be done at a faster rate to suit video recording specifications, then A/D conversion isn't as good, so an additional gain stage to scale the signal in front of the A/D helps mitigate the more noisy A/D. Depending on how good is the A/D used in K1 image sensor, it's not proven at all that the A7RIV sensor with dual gain will give better still image noise and dynamic range. In addition, on those FF cameras, the noise from A/D quantization is >100 times smaller than photon noise (more than 20dB difference between photon noise and A/D quantization noise), adding a gain in front of the A/D will be insignificant with regards to dynamic range (something negligible like 0.1ev). Although dual conversion gain may be use by marketing as a selling feature, even if dual gain will make zero practical difference in images.