What I learned from Gollywop -- and what I wonder

Started Mar 26, 2013 | Discussions thread
Detail Man
Forum ProPosts: 15,000
Like?
Re: GH2 Information
In reply to Anders W, Mar 27, 2013

Anders W wrote:

bowportes wrote:

Michael Meissner wrote:

richarddd wrote:

Another good source of information is this thread http://forums.dpreview.com/forums/post/42595072

See the links in the first post. You may have to click on the camera name in the right pane. The linked chart clearly shows the noise issue on the E-M5 between ISO 200 and 400.

Various posts over the years have said that the Panasonic sensor used by the earlier Pens tends to have less noise when you use the full stop ISO's (200, 400, 800, 1600, 3200) rather than the intermediate versions. There is an option to only use whole stops in auto ISO. It sounds like the same is probably true of the Sony sensor used in the E-M5, E-PL5, and E-PM2.

I may remember poorly, but I think I read once (somewhere in the forum) that 160, 320, 640, 1280, etc. Were the appropriate full stops on the GH2. And I had inferred from that that those were the ones to focus upon on the G5. If this is wrong, I'd like clear word on that

You are largely (though not exactly) right about the GH2 whose only "real" ISOs are 160, 320, 640, and 800.

I don't think that it is clear that ISO=800 on the GH2 is unmanipulated. See bg2b's OP here:

Real ISOs are probably just 160, 320, and 640. 200, 400, and 800 show odd artifacts (gaps or near gaps) in the RAW histograms. 250, 500, and 1000 show them even more. Everything beyond ISO 1000 is completely artificial; they're obviously just digital scalings of lower ISOs.

http://forums.dpreview.com/forums/post/37785576

This is kenw's RMS Read/Dark Noise data:

1/4000 NR-2:

160 0.73554
200 1.117
250 1.5487
320 1.3875
400 1.8514
500 2.4756
640 2.6903
800 3.513
1000 5.4109
1250 5.7917
1600 7.9696
2000 13.557
2500 13.168
3200 16.576
4000 25.919
5000 28.305
6400 35.017
8000 40.398
10000 54.246
12800 68.36

http://forums.dpreview.com/forums/post/41131538

I calculated the number of EV that certain ISOs (from rated ISO=160 through ISO=800) deviate from a straight line projection (in EV) from that which would be derived from multiplying the Read/Dark Noise data by the ratio of the Saturation ISO divded by Saturation ISO=167.

Here are results using DxOMark Saturation ISO (extrapolated SatISO values for ISO=320 and 640):

Saturation ISO -------- Read/Dark Noise deviation in EV

167 ------------------------------------------- 0.0000

209 ---------------------------------------- + 0.2791

334 ----------------------------------------- - 0.0844

413 ---------------------------------------- + 0.0255

668 ----------------------------------------- - 0.1291

However, ISO=800 shows signs of some sort of digital manipulation (perhaps of a nature other than digital scaling, as speculated by Franksson later in the same thread) which looks more like ISO=400 and ISO=200 than it looks like ISO=160. These histograms are from the DPR SCT RW2s:

http://forums.dpreview.com/forums/post/40120298

... ISO800 is the last "real" ISO, the higher ISOs are digital gains.

The measured Read/Dark Noise at ISO=800 above (relative to ISO=160) looks good (as compared to ISO=400 and ISO=200), and kenw opined that ISO=800 looked better than ISO=1000:

817 ----------------------------------------- - 0.0347

The intermediate ISOs between these are accomplished by means of digitally scaling the RAW values upwards rather than by analog amplification. The same is true about any ISO above 800. It follows that using any ISOs other than those I listed will not bring you any benefits above using the next lower ISO among those listed. They will be the same for read noise but worse for the risk of highlight clipping.

Whether the G5 behaves the same way, I don't know. And I am not sure anyone else here does either at this point. You can find out for yourself by using RawDigger and inspect the histogram for different ISOs. If digital scaling is employed, there will be gaps in the histogram (i.e., certain specific ADU values will not occur). See here for example:

http://forums.dpreview.com/forums/post/37798663

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
BothNew
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow