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

[sletts02]

Maybe not the best test when it comes to dual iso but a little comparison between 3x cropmode and regular raw using a corrected 8mm peleng and a 20-35mm wideangle shot with dual iso.

Looks promising! Unfortunate that you didn't have a contrasty blue sky to get a good feel for the range. If possible, would you be able to do a quick test where you shoot the interior of a room with windows so we can see how it handles the typically blown highlights from windows? That would be ultra helpful please and thank you 

EDIT: Just what I was after - Thanks heaps! I'm buying the 5D3 immediately now haha. It's just a shame that shipping for CF cards will take a week or two from the states.

[Canon eos m]

Nice balcony and neighborhood 

[Greg]

5D III:
5796 x 3870 = 22,4MPx
3870 / 2 = 1935
2892 x 1935 = 5,6MPx Ultra High Quality

[a1ex]

Well, it's not that easy.

In many practical cases you won't notice the detail loss (this happens when the aliased areas are out of focus backgrounds, or when the missing data is guessed correctly).

In some extreme cases (brick walls maybe?) aliasing might be so bad that it will be visible at 720x480 too. But the same is true about 1080p video with line skipping.

Anyway. The latest updates have a heuristic that I believe it puts a limit of how bad the aliasing can get (it was the key to solving the palm tree shot):

Code: [Select]

                /* if the brighter copy is overexposed, the guessed pixel for sure has higher brightness */
                fullres[x + y*w] = f < white_darkened ? f : MAX(f, dark[x + y*w]);

Anyone with good math skills can come up with a proof?

[makpat]

Hi,

I was playing a bit around Dual ISO with my 7D and took few shots.
Made DNG with cr2HDR. But somehow dont know the workflow ahead as if I open the dng in ACR it shows very dark like 100 iso shots.

Sharing my CR2 as well as DNG folder link here if someone can guide how to achieve the finest dual iso process.
https://www.dropbox.com/sh/xz77sv2bfow98s4/dZZmUIMnkZ

[ted ramasola]

@A1ex.

Since you have the cr2 of the test I did, you can check yourself what I noticed on the latest experimental you posted.
I compared it with the last 3 prior versions and I noticed that maybe your "hot pixel fix" might be getting too aggressive already as
I am now beginning to see dark spots coming out in the shadow areas which I havent seen in the previous cr2hdr builds. Check in the area I circled.
This pic I'm posting is already compressed so maybe you can verify with the cr2 I sent.

[a1ex]

These are cold pixels and they are no longer corrected. I'll take a look.

[a1ex]

I think I've got it working on 5D2: https://bitbucket.org/hudson/magic-lantern/commits/b3999097eed7

Also, since the experimental converter handles almost all cases better than the old one, I've updated the converter from the first post (so cr2hdr.exe now has the latest algorithms).

[batca.dragos]

So does the 5D2 have a 8-channel readout after all a1ex ?

[a1ex]

It doesn't need an 8-channel readout, but two separate analog amplifiers. The paper says:

Unfortunately, most other Canon cameras follow a different pattern for the CMOS #0 register,
without any obviously duplicate fields.
...
We can now speculate that cameras with 8-channel readout have two separate amplifier circuits,
and you can program the ISO for each one separately.

Translation: at that time I knew how to program the cameras with 8-channel readout (because it was obvious) and I had no idea how to program the others, or whether they will work or not. Now I've discovered the missing bit from 5D2 and it works.

I still have no idea about the other cameras. I'd say the 6D has 99% chances to work, since 1% almost got it working, but I did not check the new trick on any other camera. Follow the ADTG thread and give it a try.

[batca.dragos]

Understood now  . Congrats on the find.

[ted ramasola]

A1ex,

is there a compiled 5d2 dual iso module I can test?

[ilguercio]

It doesn't need an 8-channel readout, but two separate analog amplifiers. The paper says:

Translation: at that time I knew how to program the cameras with 8-channel readout (because it was obvious) and I had no idea how to program the others, or whether they will work or not. Now I've discovered the missing bit from 5D2 and it works.

I still have no idea about the other cameras. I'd say the 6D has 99% chances to work, since 1% almost got it working, but I did not check the new trick on any other camera. Follow the ADTG thread and give it a try.

50D has got 2, from my pictures.
Right?
http://chdk.setepontos.com/index.php?topic=6161.180

[1%]

50D, photo mode:

Code: [Select]

/* 00:00:04.078841  */ CMOS
/* 00:00:04.078911  */     00 0004 404B548E

/* 00:00:14.214297  */ CMOS
/* 00:00:14.214376  */     00 0094 404B549C

/* 00:00:26.551046  */ CMOS
/* 00:00:26.551116  */     00 01B4 404B54AA

/* 00:00:47.349123  */ CMOS
/* 00:00:47.349194  */     00 016C 404B54C6


Code: [Select]

       PHOTO_CMOS_ISO_START = 0x404B548E; // CMOS register 0000 - for photo mode, ISO 100
        PHOTO_CMOS_ISO_COUNT =          5; // from ISO 100 to 12800
        PHOTO_CMOS_ISO_SIZE  =         14; // distance between ISO 100 and ISO 200 addresses, in bytes

        CMOS_ISO_BITS = 3;
        CMOS_FLAG_BITS = 2;
        CMOS_EXPECTED_FLAG = 0;
is_5d2 = 1;   
 
This produces Iso1 != iso2 right after its set. Using size of 12 causes err reg not 0. Setting cmos expected flag like 5DII errors on the flag.

*HiNT. Seems the addresses when cmos updates all regs together are not what we're looking for... its when reg 0 is updated by itself.

[a1ex]

For 50D, these are the values for ISO 200/400/800/1600?

First we need to know what bits do what. You can set the values manually, e.g. *(uint16_t*)0x404B548E = whatever. Only after you are sure about the meaning of each bit (at least the two ISO fields), you can configure the constants in dual_iso and see the errors.

For experimenting, just change the memory values manually.

[1%]

On 50D they change on the middle ISOs.. like 160, 320, etc. I think somewhere at the begining of the thread I did what went with what. 04 is 100-160 tho. I went 160, 320, 640, 1250. I think I didn't do the last one.

4 I think is only 100
94 is        100 10 100
1B4         110 110 100
16C         101 101 100
So which ones are likely to be ISO.

[Elmeri]

could it be
004          000 000 100
094         010 010 100
1B4         110 110 100
16C         101 101 100
so is it       iso iso xxx

[ted ramasola]

Is it ok to report tests and bugs on dual iso for 5d2 here? or should I start another thread for 5d2 dual iso?

did some tests on the dual ISO and heres some quirks.

1. after activating dual iso, i can take 1 shot with it then the following shots are normal cr2. I did not turn on alternate.
I sometimes take the card out and back before I can take a dual iso shot. Sometimes reboot cam will work sometimes not.
or I will download the cr2 via usb cable. after downloading i will take another shot, they will not be in dual iso. looking at ML menu dual iso is still at ON.

2. changing the recovery ISO has a bug, after entering submenu i can change it to 3200 for example. exiting sub menu I can see the the dual iso reads 100/1600 still.

I have no problems using dual iso with my 7D.

[1%]

6D photos, all ISOs starting with 100-128k.. there is no 66?

03 0000 40450E08
03 0011 40450E1A
03 0022 40450E2C
03 0033 40450E3E
03 0044 40450E50
03 0055 40450E62
03 0077 40450E74

100 0 1
100 0 10
110 0 11
100 0 100
101 0 101
111 0 111


Seems like ISO 0 ISO

50D seems ISO ISO 100

so how to set up flag bits, etc for this.. also on 50D the reg address never changes in LV.

[Audionut]

Also, since the experimental converter handles almost all cases better than the old one, I've updated the converter from the first post (so cr2hdr.exe now has the latest algorithms).

The cr2hdr in the first post is timestamped 30/07/13?

[Danne]

Hi @Alex! Tried some dualiso and some birds. The converter seems to have trouble converting certain lighting conitions such bright backgrounds with smaller subjects lika flying duck for instance? The first converter managed to convert this but not the latest

https://docs.google.com/file/d/0B4tCJMlOYfirWmVDd1pZVlYwU2M/edit?usp=sharing

Is the conversion dependent on motif? Have some pictures of an eagle , rather tiny in the original but could be cropped after conversion if converted. Seem the next latest of the converters could do some of the files the other two converters can,t handle but not all of them.

*update. Added the exiftool, didn,t think it was necessary but seems to convert ok now, still not able to convert linked example though.
Another strange thing is that the converter creates an extra dng_original file when exiftool included? Exact same size as the converted dng-file.
*update2. The converter gives random results. Sometimes it doesn,t recognize the cr2,s as interlace iso, but retrying, suddenly it works?



Thanks
//D

[Canon eos m]

Is there crop mode functionality on the 5DM III shooting video?

[mauerfuchs]

Whats the difference to just use the RAW function in 50fps mode in combination with the older HDR function that records two different frames. I mean...wouldn't it help the antialiasing a bit? The motionblur now may be the new old problem...
i got a 5d mark III.

[swinxx]

hello ml team!

is it possible to name the dual iso shots automatically (e.g. DI_99A6430.CR2 .. because in the mac finder it is not possible to see which cr files are dual isos and which are not.. ?

a really satisfied ml user.. 

[a1ex]

Wow, not a single sample from 5D2?!

See you in roughly 2 weeks (maybe earlier if I get net access).