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

Started by jcdenton, July 28, 2014, 10:40:10 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

jcdenton

For one year I owned Canon 50d and shooting a lot of cool videos using RAW module. Recently bought brand new 5d mk3 and installed a Magic Lantern.

Recording goes ok (MLV + Audio) but when I copied, converted and open files in ACR noticed that they are 1 stop darker then they should be. Especially considering that they must be overexposured a little beacouse I was using a Histogram ETTR RAW Hint. It just 1 stop darker compered to h264 and raw_rec with the same settings and static conditions.

Here are some details what I found out:
- I tried converting MLV to dng using MLV Mystic (both Win/Mac) and MLV Converter 1.9.2
- I tried 1.1.3 and 1.2.4 firmware (clearing the bootflag everytime)
- I tried Nightly.2014Jul25.5D3113 and Nightly.2014Jul16.5D3123
- Overall exposure on the camera monitor seems to be the same. Result are different only on computer.
- Raw rec & h264 have correct exposure as it should be.

So for now nothing helps and I shooting using the raw_rec. But the ability to record sound in the same files using mlv rec are very useful and I'm looking forward to find out what is the problem.

Have someone any ideas why is it  happening?

jimmyD30

I experienced the same thing, for now I'm using canon exposure for shooting mlv and all is fine.

a1ex

Proof please (upload two DNGs to show the difference).

jimmyD30

I can do some testing later today, can you suggest a method which would help ensure valid results?

a1ex


jimmyD30

Great, I'll also take some pics of the LCD to show exposure details.

jcdenton

I'll try to post some test examples later this evening.

jimmyD30

Ok, just did an indoor test and I'm not really seeing a problem :D

In the past I was doing outdoor shots and I'm thinking now the extreme of the skies may have been pushing the histogram further to the right, having me adjusting exposure down.

Outdoors is very overcast today, so won't be able to do additional experiment for now :(

jcdenton

Didn't get a chance to post video examples. Problem is still there. I'll post it today.

mannfilm

Canon 5dm3, multiple builds and converters. This has been and is happening with me. First noticed it 3 months ago when doing tests to decide to change from RAW to MLraw. Real noticeable with exteriors.



a1ex

Can you use the reference converters? (raw2dng and mlv_dump)

You have used closed source software, so I have no idea what processing was applied.

jcdenton

Quote from: a1ex on July 29, 2014, 08:39:54 PM
Can you use the reference converters? (raw2dng and mlv_dump)

You have used closed source software, so I have no idea what processing was applied.

a1ex, I already tried raw2dng, result doesn't differ. Also I will post source here, so everyone can try their converting methods, but I'm think it wont help.

https://www.dropbox.com/s/vuf7qv47sxjigrr/M29-2115.MLV (639MB)

https://www.dropbox.com/s/y3pd4i2lspfyuuz/M29-2115.RAW (563MB)

a1ex

It does differ; raw2dng does not output 16-bit files, like the one you attached.

I'm not going to download these huge files.

jcdenton

Quote from: a1ex on July 29, 2014, 08:49:14 PM
It does differ; raw2dng does not output 16-bit files, like the one you attached.

I'm not going to download these huge files.

ok I'll post it in a few minutes.

jcdenton

Quote from: a1ex on July 29, 2014, 08:49:14 PM
It does differ; raw2dng does not output 16-bit files, like the one you attached.

I'm not going to download these huge files.

Well. It's really does differ. My mistake :-[

MLV (mlv_dump):
https://www.dropbox.com/s/etr4fu3szb37vna/mlv_mlv_dump.dng

RAW (raw2dng):
https://www.dropbox.com/s/g2do4cmxglm9p4b/raw_raw2dng.dng

But It's even darker.. I'm not sure is this right because of the fact that histogram on the camera doesn't match the result histogram and I can say it's not match even visualy. How to set the right exposure then. I'm confused.






a1ex

Here, mlv is rendered a little brighter, and so is its raw histogram, and the spotmetered area.



Metadata: both files have white level at 15000 and black level at 2048.

There must be some other metadata interpreted by your program in a different way.

One difference is AsShotNeutral, but recent versions of raw2dng use the same value as mlv_dump. Make sure you have a recent raw2dng - first post at http://www.magiclantern.fm/forum/index.php?topic=5404.0 .

If it's not that, you'll have to compare the metadata of both files, and find the tag that causes the exposure difference. Or just drag the exposure slider a little higher.

jcdenton

a1ex, I'm not sure that the problem is with the tiny difference between mlv's and raw's exposure. But the problem is with the exposure at all. As I see you had to use a 2 extra stops to properly exposure my images that was already properly exposured in the camera according to the exposure meter. It's mean that every shot that was made on 5d mark III need to be corrected. The 50d didn't have such a problem and the output images looked exactly as they was shot, with the ETTR hint all you had to do is just slide the highlight recovery all the way to the left get it back to normal and get your dynamic range with no noise. Now its not that easy and I'm trying to figure out how to use a ETTR hint now.

a1ex

Your ETTR hint indicates 1 stop of underexposure, and ETTR itself uses another stop as a safety margin (the algorithm needs it to know how much to go back, but you can reduce it to 0.5, or allow it to clip some more highlights, or ignore some color channels).

Your issue was about the difference between raw and mlv (read the title of your post if you don't believe me).

jimmyD30

I can tell you I didn't read the title of this post properly ::) and initially thought we were talking about how using the histogram and ETTR/ETTR hint is causing under exposure or at least under exposing compared to the Canon exposure meter, which I (and a few others) seemed to have been experiencing (for me appeared to be about 1.5 to 2 stops).

So thank you for the explanation of how ETTR/ETTR hint works, now I can see why this might be the case, even though the exposure discrepancy seems to be more prevalent outdoors.

a1ex


jimmyD30

Perfect! This further explains my discrepancies outdoors with bright skies in the upper part of my framing :D

So ML ETTR is doing what it's supposed to taking the whole scene into consideration, rather than Canon's mostly center frame metering. Make sense now.

I can see where depending upon circumstances and preferences one method would be desired over the other (using Canon metering vs ML histogram and ETTR).


jcdenton

Quote from: a1ex on July 30, 2014, 07:48:22 AM
Your ETTR hint indicates 1 stop of underexposure, and ETTR itself uses another stop as a safety margin (the algorithm needs it to know how much to go back, but you can reduce it to 0.5, or allow it to clip some more highlights, or ignore some color channels).

Your issue was about the difference between raw and mlv (read the title of your post if you don't believe me).

a1ex my apologize for naming the post wrongly and for some misunderstanding. I'm just trying to understand what is going on and why the canon exposure meter don't match the result on the computer. And how should I setup the camera to achieve ETTR advantage.

As I understand your advise is to first exposure according to ettr raw hint and then give it an extra 0.5 stop to compensate underexposure in post?

As I understand you said that the problem of underexposured image in mlv is the wrong interpretation of meta data by ACR and not the raw files themselves? So the files are exposured right and they have entire Dynamic Range according to histogram in the camera. They just underexposured in ACR and need to be corrected?

Sorry if I get you wrong as my English is not very good. And thank you for your help!

mannfilm

Not using ETTR. MLraw files are not exposed right, they are under-exposed, and do not have the same dymanic range as raw. This is not ACR meta-data whatever. Maxing the shadows in ACR reveals that MLraw dymanic rane is permanently under-exposed. MLraw shadows detail cannot be lifted to Raw levels. I have to over-expose by 1.5 stops to capture the same dymanic range as raw or canon native.