A7*2 Resolution Drop with >3.2s Exposure

[thrice]

I imagine quite a few of you have seen the Petapixel article about the 'star eater' behaviour in Sony A7RII 3.3 (and A7SII 2.10) firmware.

I have also observed another undesirable attribute of this behaviour - it destroys resolution.

All taken with Pentax HD 35mm adapted to Sony. Camera set to Uncompressed RAW. All NR/Steadyshot settings off. Remote release and MF used for all images.
All settings between the images identical except for shutter speed and exposure adjustment in post.

I shot the images 2 stops apart, pushed the 1.3s exposure up by a stop, and pulled the 5.0s exposure down by a stop.
Original RAW of 1.3s image: https://www.dropbox.com/s/p050wp1wqof5nb5/DSC00139.dng?dl=0
Original RAW of 5.0s image: https://www.dropbox.com/s/i4d6pdpsd9g6bzm/DSC00140.dng?dl=0

Full Image:

1:1 no sharpening

1:1 with high sharpening

2:1 with high sharpening

What you see is the squiggly watercolour effect that you usually get with high levels of noise reduction, there is also a noticeable drop in metamerism (colour differentiation).
The same effect can be seen to a higher degree on images with finer grain (ie. rocks and sand).

I have another example image here which displays very low resolution and micro-contrast because most of the detail is pixel sized and Sony has decided I don't need that.
It wasn't a controlled test so unfortunately I don't have a shot of the same subject with a shorter exposure, so take it with a grain of salt.
Full Image:

No Sharpening

High Sharpening

 

[thrice]

This makes it a little clearer.

 

[Vivek]

This is a fair weather "all rounder ".

Also, it is a bit dated. What measures (increasing the ISO, for example) have people taken to mitigate this problem?

I decided to skip the gen II cams altogether. Although, I might pick one up when the price drops to my figure.

 

[thrice]

This is default spatial filtering on images with exposure longer than 3.2s

It is not possible to work around or disable. It affects A7Sii and A7Rii, it might also affect other cams.

It was well known to kick in when using bulb or longer than 29s exposures but with the latest firmware it kicks in when exposure exceeds 3.2s.

Sony obviously think it's a good idea so whatever camera you buy from them next will likely have the ‘feature’ unless they're discouraged.

This is a fair weather "all rounder ".

Also, it is a bit dated. What measures (increasing the ISO, for example) have people taken to mitigate this problem?

I decided to skip the gen II cams altogether. Although, I might pick one up when the price drops to my figure.

 

[jdphoto]

When I owned the Sony A7rii, I upgraded the firmware to mitigate the ODS delay when using the Elinchrom ELB 400/HS head. I decided to sell it when the talk of of a new model with two SD cards was rumored. I wanted a jump on the flood of second hand Sony's before the price drop, so didn't experience long exposure issues as I hadn't used it in that regard yet. The loss of micro contrast in anything longer then 3.2 seconds is absurd. The A9 however, has not piqued my interest and the launch price confirmed that. So for now i'll wait on Sony. This is something Sony needs to fix. I like long exposure photography and won't consider their cameras until this is fixed with firmware. Thanks for the article.

 

[Vivek]

This is default spatial filtering on images with exposure longer than 3.2s

It is not possible to work around or disable. It affects A7Sii and A7Rii, it might also affect other cams.

It was well known to kick in when using bulb or longer than 29s exposures but with the latest firmware it kicks in when exposure exceeds 3.2s.

Sony obviously think it's a good idea so whatever camera you buy from them next will likely have the ‘feature’ unless they're discouraged.

This is acknowledgement from Sony of their hardware limitations.

The question however is about the work around. How about keeping the exposures below 3.2s while upping the ISOs? Would that yield better clarity ?

 

[pegelli]

Hi thrice,

I downloaded both dng's and inspected them in Lightroom myself at all kinds of magnifications.

Maybe I'm looking wrong, but I really can't see a resolution difference between the two. The only conclusion I would draw is that the 1.3 sec exposure has a tiny bit more noise vs. the 5 ssec exposure and therefore it seems sharper to the eye at 1:1. However when you blow up further (see my example at 8:1 below) you'll see the details seem just as large, only the 1.3 sec exposure sometimes gives a few brighter pixels making you think it's sharper but I feel it's just a little bit more variation due to more noise. However the details are just as wide. So the microcontrast of the 1.3 sec exposure indeed seems a wee bit higher, but it might be microcontrast that isn't really there.



Hope this helps

 

[Vivek]

Dan, perhaps there is a chance that sony software folks look at how things look when an image is looked at larger than 100% magnification to justify their crippling the capabilities of their sensors.

It is also possible that the "star eater" issue is due delusional amateurs imagining things which aren't there at all?


Or


May be we need to get this EXMOR R sensor in a Pentax camera or even a Nikon one to get the best outputs...

 

[thrice]

Pegelli I definitely see the resolution drop even on the very selective crop you made. Have a better look around the image.

Vivek, even Jim Kasson has confirmed it with very scientific testing so us amateurs really should defer to his expertise.

 

[Malina DZ]

I have also observed another undesirable attribute of this behaviour - it destroys resolution...

Thank you for your insight and providing DNGs to support your claim.
I compared both files and can notice some chroma noise and grain on a 1.3sec exposure shot that is noticeable once you zoom in past 100%. 5.0sec exposure has no noise at all. Obviously, in-camera processor kicks in a NR algorithm to clean it up and a minute fine detail suffers as a result. I would not call it as strong as "destroys". That iso100 noise creates an illusion of an extra detail.

 

[thrice]

Might be the additional noise from pushing 1 stop vs pulling 1 stop.
Essentially there should be 2 stops additional dynamic range in the shadows of the 5.0s image.
Even adding NR in Lr (10 luminance 25 chroma) to the 1.3s image to equalise the noise I see more detail in the 1.3s image.

Since this is erasing spatial frequencies at or close to the nyquist limit, it is essentially only affecting higher MTF frequencies. If you are happy with cheap glass then you probably won't care about this.
However if that is the case you may as well save some $ and get the A7II instead of the A7RII.

 

[pegelli]

Pegelli I definitely see the resolution drop even on the very selective crop you made. Have a better look around the image.

You're right, after a good night sleep and a good look around I can see loss of very fine detail that's not noise. I shouldn't have made comparison after a hard days work just before going to bed

I'll send an email to the address in the Petapixel article. In the end Sony provided lossless raw after a lot of users outcry, so using the same technique they might put the time limit for this back at 29 sec (or longer) or even better provide a menu option to turn it off, so the user can decide for himself if he wants it or not.

A question for you: do you see any difference of this phenomena with Long Exposure noise reduction on or off?

 

[thrice]

A question for you: do you see any difference of this phenomena with Long Exposure noise reduction on or off?

Hi Pegelli,

The guys on the lonely speck and cloudy nights tested every possible combination of NR setting and file encoding and found no way to avoid the behaviour. It appears baked into the RAW regardless of settings.
I have the 3.20 firmware and have found a way to unpack and change the firmware version to avoid it being rejected as an update to 3.30. Sadly I have not found a way to re-pack it in the limited time I've had to play.

Also, it's probably better to wait and see if Sony come to the party. If they stick to their guns I might look at the pentax K-1.

 

[Vivek]

I would urge Sony to offer their EXMOR R and EXMOR RS sensors to Pentax and others!
:lecture:

 

[jdphoto]

Does this affect other Sony A7II models?

 

[Malina DZ]

Might be the additional noise from pushing 1 stop vs pulling 1 stop.

I actually pulled exposure by -2.0 EV on a 5sec shot and left all values for 3.2s one at 0, so there is some inherent noise at iso100 without pushing the file.
BTW, was silent shutter/EFCS mode engaged when these shots were taken?

 

[thrice]

I actually pulled exposure by -2.0 EV on a 5sec shot and left all values for 3.2s one at 0, so there is some inherent noise at iso100 without pushing the file.
BTW, was silent shutter/EFCS mode engaged when these shots were taken?

EFCS. Of course the longer exposure will have less noise, it has 4 times as much light!

 

[thrice]

the 'Sony listens' e-mail address responded with "Call our alpha support line".

 

[Vivek]

the 'Sony listens' e-mail address responded with "Call our alpha support line".

This thread will make them listen. I think they do as well.

Luckily, it is not sabbotaged and locked.

 

[thrice]

I rolled back to 3.20
Sony don't seem to be doing much about this.