5D mk3 mlv_rec exposure level -1 stop (raw_rec is ok)

[a1ex]

Proof please.

[jcdenton]

Have some tests.

Same setting, same condition. MLV vs CR2.

MLV

https://www.dropbox.com/s/uc5g48qshid0twc/M31-2250.000000.dng

CR2

https://www.dropbox.com/s/w1s9pkxjmgre45r/2C1A0087.CR2

0,75 stop difference.

[jcdenton]

Took mlv test shot to same exposure as CR2 by moving the exposure in ACR. And compered noise level.

Croped 250x250px and 200% resize.

CR2


MLV


Resized to 1080p. Croped 250x250px and 200% resize.

CR2


MLV


When MLV frame is formed in the camera the 5760x3840 resolution resized and croped to 1920x1080? So the noise level should be much less then in CR2. Didn't it?

[a1ex]

LiveView ISOs are a little lower than displayed. I've measured 82 on 5D3 LV, vs 104 in photo mode - that is, 0.34 stops difference.

You took the CR2 at F1.4 (from exif), and at this aperture, Canon applies 0.15 stops of digital gain (found by DxO and confirmed by my experiments - the exact value is log2(0x239/0x200) from the digital gain configuration).

So, you should expect the MLV to be 0.5 EV darker than the CR2, assuming identical metadata.

I've converted your CR2 into DNG with Adobe DNG converter, so ufraw renders it like any other DNG (that is, normalizing the raw data according to black and white levels from metadata, and not other black magic), and I still needed 0.45 stops of exposure compensation for the MLV.

   

Noise levels are obviously going to be different (see the first link). The 5D3 does hardware 3x3 binning on the sensor, and then it adds read noise from its electronics (after binning). The end result is worse than a CR2 downsized by 3x3 in software, both in sharpness and noise levels.

=> this is not a bug in MLV.

[jcdenton]

LiveView ISOs are a little lower than displayed. I've measured 82 on 5D3 LV, vs 104 in photo mode - that is, 0.34 stops difference.

You took the CR2 at F1.4 (from exif), and at this aperture, Canon applies 0.15 stops of digital gain (found by DxO and confirmed by my experiments - the exact value is log2(0x239/0x200) from the digital gain configuration).

So, you should expect the MLV to be 0.5 EV darker than the CR2, assuming identical metadata.

I've converted your CR2 into DNG with Adobe DNG converter, so ufraw renders it like any other DNG (that is, normalizing the raw data according to black and white levels from metadata, and not other black magic), and I still needed 0.45 stops of exposure compensation for the MLV.

   

Noise levels are obviously going to be different (see the first link). The 5D3 does hardware 3x3 binning on the sensor, and then it adds read noise from its electronics (after binning). The end result is worse than a CR2 downsized by 3x3 in software, both in sharpness and noise levels.

=> this is not a bug in MLV.

Yep. I think you are right.

I have a manual lens with chipped m42-eos adapter. This might be the problem? What aperture should I select so the canon apply no digital gain to the image? Taking the opportunity to ask you. How can I enable the 66 - 115 - 230 etc. iso on 5dmk3? I only have 100 - 160 - 200.. Thank you for your help!

[a1ex]

From the linked graph, f4.0 and beyond does not boost the digital gain.

The module for low ISOs is not ready yet, but on 5D3 you can compile ML yourself and apply the tweaks manually with the iso_regs module. On other cameras, you can apply the tweaks with the adtg_gui module.