Dual ISO - massive dynamic range improvement (dual_iso.mo)

Started by a1ex, July 16, 2013, 06:33:50 PM

Previous topic - Next topic

0 Members and 7 Guests are viewing this topic.

ted ramasola

For the 3 images and choosing with a relaxed eye, and not scrutinizing/pixel peeping the image,  C is more "pleasing" than the rest, its less noisy.

It might not be the same choice with a different subject matter and composition.
5DmkII  / 7D
www.ramasolaproductions.com
Texas

mannfilm

Read the PDF, maybe I missed this. Regarding 50% vertical rez loss in dark and highlights. So on either end we're getting line doubled 1920 X 540  instead of true 1920 X 1080? And whats the range of the lower rez? If we are getting 14 stops, is the low  rez sections in the bottom 4 stops and upper 4 stops or what? You guys are the new gods of film. 

a1ex

Figure 8 shows where you get full resolution.

With my latest modifications, you also get almost full resolution in deep shadows, but with more noise (see my previous post).

driftwood

Canon 60D, Canon 5DMK3, Lexar 1000x 128GB CF, Panasonic (shhhh!) GH2s & GH3s. :-)

Toffifee

I'd pick picture C, but I could go B.

C obviously has less noise, it also doesn't seem to soften up the details.
Aliasing seems about the same, A a tiny bit better, but it's really noisy compared to the other two.

Audionut

C because it has the least noise and artifacts.  Although it has more aliasing around the top silver edge of the camera on the table and the table edge, and more color noise in the silver edges (of the camera).  There is more perception of detail in the table in B, but it looks to be the luma noise.

A has by far the least aliasing.

MA Visuals

Quote from: a1ex on July 18, 2013, 05:39:52 PM

@MA Visuals: can you repeat your test with these 3 raw2dng's? Just post-process the same video with them.


Rendering now... will upload the results shortly.

driftwood

median6 v mean23 v shadow25 v shadow 50 v shadow 100

https://www.dropbox.com/s/noa57vx3h6g7r0y/shadow25_50_100test.png

I really like shadow 50 together with the less purple artifacting in mean23
Canon 60D, Canon 5DMK3, Lexar 1000x 128GB CF, Panasonic (shhhh!) GH2s & GH3s. :-)

a1ex

Nice, which one do you like from Driftwood's test?

Another tip: in crop video mode, it will skip only 2 lines at a time (instead of 6). Therefore, you can expect a lot less aliasing.

You can also record at higher vertical resolution (1320) and then resize it to something lower. It should hide some artifacts.

ted ramasola

my choice would be either 25 or 50.
50 has less aliasing.
25 has less noise

5DmkII  / 7D
www.ramasolaproductions.com
Texas

ajay

Quote from: a1ex on July 18, 2013, 05:39:52 PM

@ajay and @audionut: can you develop your test shots (the same pictures that you posted earlier) with latest cr2hdr?

A1ex ... I took the two images that I sampled before using the latest cr2hdr and the halo effect is completely gone from the orange. The second image still has moire caused by the feathers, but the false colors are reduced quite a bit. Definitely better with both images.

AJ

GregoryOfManhattan

have tried out the dual_iso module with raw2dng at 7351be3d3c0c
sequences processed through resolve.

room was so dark i moved the Base ISO up to 800 and that may not have been enough.
did no camera preparation before this event - so i can't confirm that the comparison clips had significant Ev offsets applied.

dual iso mode seemed to work at all resolutions i tried - here's a 3586x1320 DNG
http://50.56.67.113/ml-17july13/000012.dng
http://50.56.67.113/ml-17july13/1927-3584x1320-000012.jpg - same thing as an RPP processed jpg at 1/2 scale



3.5K did not seem to work as well as builds from last week or at least the usual small and memory hacks didn't work as well when i tried them - only got 1-2 seconds of 3.5K.  2.2K seemed continuous.

had one lock up where i needed to pull the battery

which repository has the code with median23 and other options? will have time to test this weekend.

a1ex

To choose other options, just flip the conditional defines from the source code (INTERP_*). For shadow aliasing, adjust f_shadow.

GregoryOfManhattan

Quote from: a1ex on July 18, 2013, 07:50:04 PM
To choose other options, just flip the conditional defines from the source code (INTERP_*). For shadow aliasing, adjust f_shadow.
great we can make our own custom blends.
modules/lv_rec/raw2dng.c
modules/dual_iso/cdr2hdr.c

if a clip is dual_iso will the current raw2dng always display the HDR info and the EV offset?

MA Visuals

Ok... here's the new comparison using shadow100, shadow50, and shadow25.  I enabled the download link on Vimeo for those wanting a more accurate representation of the footage. 

Shadow 50 has my vote as a set it and forget it default. 

Shadow 100 is too noisy... Shadow 25 has too much aliasing for me to use the footage... Shadow 50 has minor aliasing but much less noise than Shadow 100.  On the other hand... Shadow 100 combined with a light pass of de-noising in post would work as well.

Vimeo link... https://vimeo.com/70574082


mauerfuchs

Is the dual iso method similar to the canon video raw strategy?
http://blog.abelcine.com/2012/11/05/what-exactly-is-canon-raw/
I bet some of the C500 developers who are watching ML Raw development remember some of these steps very well.

driftwood

A look at the artefacts of each of the test interp modes. Same dual ISO DNG opened 'As Shot - 5500 / Tint +17' (RawMagic and v0.13 OSX raw2dng no interp modes and opened with a tint of +66 then changed to +17).

https://www.dropbox.com/s/a3wd7s9sh5ud2sv/interp_artifact_test.png
Canon 60D, Canon 5DMK3, Lexar 1000x 128GB CF, Panasonic (shhhh!) GH2s & GH3s. :-)

JackDaniel412



Dual ISO test on 5D3 (ISO 100 & ISO 1600), post-processing using "raw2dng_mean23" and After Effects.

portmantoad

ok so I have a few thoughts here.

when using this in a video mode (60p as an example), could you alternate which exposure you use each row for each frame? that way you'd have the full resolution midtones output as it is now, as well as what is essentially a iso100 60i output and a iso1600 60i output. I would imagine losing some temporal resolution is probably better looking than losing half of your vertical resolution.

the other idea is, although I have much less hope for this to be feasible, is it required to choose the exposure per row or can it be adjusted on an individual photosite basis? Because if that's the case we could take a page out of fujifilm's X-Trans filter strategy and set up the sampling in a pseudo-random fashion. That would make the aliasing less obvious and could actually be kindof cool and filmic.

a1ex

I can alternate the exposures, but the postprocessing will be a lot more complex. It should help with highlight aliasing (because the shadow one is pretty much solved).

In some videos, the exposure pattern is moving, and in others it's fixed. I don't know yet why this happens.

I don't know how to alternate ISO for every single line - if you look in the paper, all I do is changing a register to low-high or high-low. The ADTG chip does the rest.

driftwood

a1ex: In the dual ISO raw2dng interpolation, what ways are you looking at improving the accuracy of returned values and reducing artefacts? Ah I see you're already answering this.
Canon 60D, Canon 5DMK3, Lexar 1000x 128GB CF, Panasonic (shhhh!) GH2s & GH3s. :-)

sarangiman

Quote from: horshack on July 18, 2013, 04:44:12 PM
For the shadows the ISO 100 lines are being discarded from the composite raw, which reduces the light-capturing surface area of the sensor in half.

Thanks Horshack-- funny how the obvious sometimes escapes me. Although, b/c they're interpolating between the two ISO exposures, you can set up the algorithm to only use the ISO 1600 file where read noise trumps shot noise contributions significantly. But I see what you're saying -- when you do that, you pay a cost due to shot noise contributions (vs. exposing the whole image at ISO 1600, for example).

But typical 'engineering' DR calculations using SNR of 1 as the lowest signal don't consider shot noise, correct? Therefore, one could still say that the engineering DR is extended to FWC/read noise @ISO 1600, which is ~14EV. If you're using a higher SNR cutoff of, say, 20, then the DR is much more modest (considering the increase in shot noise contributions you mention).

MA Visuals

I updated my original test to use the optimized version of RawToDNG (raw2dng_shadow050.exe).  Aliasing is noticeably reduced now. Thanks Alex.


dubzeebass

I'm finding it difficult to properly set exposure, though I'm sure this will improve as code revisions progress.  This RAW converted with the Shadow 50 executable is particularly special.


jkdjedi

Sorry if I missed it...but....Is there a noob guide on how to implement this hack on the 7d? My guess is that you'll need the original Magic Lantern firmware and just copy and paste to some bin files..??? I know, I know...we're not worthy..but please, point us in the right direction..Thank You. :)