A Close Look at the Gradability of H264

C

Chris_TC

Active member
I love reading about HDSLR cameras on various forums. And the amount of FUD I encounter often baffles me. One of the things you will most frequently hear is how much the H264 codec of 5D and 7D sucks and how you can't grade it.

We all know that you can't grade 5D/7D footage to the extent you can grade Redcode RAW. But way too many people outright refuse to even look at the facts. Most of the people who shoot down the codec haven't ever tried to push it in post. So here we go.

I don't usually grade to an extent at which the grade turns into a special effect. I like to keep it subtle. But in the below example I changed the entire mood of the scene significantly. So it should be an interesting example.

Some quick info on the clip itself: it was shot on the 5D with the 24-105, sharpness turned all the way down, locked off on a tripod.

Let's start with the full resolution 1080p images for your reference:
Unaltered
Graded
Sharpened
Grain Added
DVD Resolution


This is how the clip looked out of camera:
g1_small.jpg


The first thing I do is grade it using a variety of filters:
g2_small.jpg

g1_crop.jpg


Next up is the crucial step of sharpening the footage. I use a two-step approach. To increase local contrast, I first sharpen with an extremely large radius and a low intensity. To increase detail sharpness, I then sharpen with an extremely small radius and higher intensity (left image is graded, not sharpened - right image is graded and sharpened twice):
g2_crop.jpg


If you take a close look at the full resolution images, you will notice that the sky shows a bit of posterization. To my eyes, this is the only visible compression artifact after this heavy grading. What I like to do is add some extremely subtle grain. It makes the image look more organic and less sterile, and it hides compression problems:
g3_crop.jpg


I also often read how 5D/7D footage is barely better than SD. So I scaled down the final image to 720x486 (SD DVD resolution) and back up to 1920x1080 as it will be displayed on a 1080p screen (left = graded HD, right = scaled down and back up):
g5_crop.jpg


And finally a comparison between the original out of camera image and the fully graded HD shot:
g3_small.jpg

g4_crop.jpg


I will not comment on those results, you be the judge.
 
It looks pretty good. The only thing I would comment on is the slight breakdown of color in the sand. Other than that i love it.
 
conversion

conversion

Have you converted the H264 to a better codec/color space before grading?

When you convert h264 to a codec with a higher color space (ProRes for example) you're grading will get better.

You really need to do this.
The quality of the source files won't get better ofcourse but the color space in which grading effects are applied does get better.

Look at this 5D example. It looks fantastic. Including the grading.
http://www.vimeo.com/6549643


:2vrolijk_08:
 
Last edited:
Paul Jay said:
Look at this 5D example. It looks fantastic. Including the grading.
http://www.vimeo.com/6549643

Your examples look awful. No offens;)
Which is probably you're own conclusion.

:2vrolijk_08:
Click to expand...
What's your point exactly? Why do you post a clip with a bunch of shallow depth-of-field closeups and crushed whites/blacks?
I picked a landscape scene with tons of micro detail for a reason, but that's apparently lost on you.

The clip you linked to neither has an ungraded version online nor a 1080p version. It contributes nothing at all to this thread.
 
The link that he posted (Camille's work) is a great example of "grading" versus your example which is more like "Correcting". Although she didn't post a before and after, I can assume that the original footage looked NOTHING like that.

Your examples show more of a simple shift in hue and color temperature. It's not an example of how the compression handles when adjusting attributes such as Highs, Lights, Mids, Darks and Shadows. Have you done an example that's aimed at a more stylized look, similar to what Camille's colorist has gone for?

Correcting footage shouldn't really be hard, but an all out "grade" is a different story.
 
Kholi said:
The link that he posted (Camille's work) is a great example of "grading" versus your example which is more like "Correcting".
Click to expand...
Any change in color and luminance is considered grading, including correction.

Your examples show more of a simple shift in hue and color temperature. It's not an example of how the compression handles when adjusting attributes such as Highs, Lights, Mids, Darks and Shadows.
Click to expand...
I've adjusted lows, mids and highs in color all separately. I've added a colored grad. I've adjusted the contrast both globally and locally. I've done all the things you'd do on a typical grading job.
In fact, I've done a lot more than you would on a typical grading job.
 
Thanks for posting Chris. I think the results look pretty good for what you did to it. Great test if you ask me.
 
I think the look and your before after comparisons look pretty great. One question though, isn't it moreso a matter of render time and not quality of look that H.264 impedes on? I thought that in a raw format, your renders on a color grade happen much faster than on a production format. If I am off base though, please let me know as transcoding everything I shoot before I edit is a step I would love to eliminate, heh.
 
As with any video codec, the more color space you have to work with, the more you can manipulate the chroma channels. This is especially important if you are interested in "boosting" the levels of chrominance, not necessarily lowering them.

Where a video codec falls apart is when you begin adjusting the saturation for example. If you increase the saturation of a video clip you will eventually see posterization. You will be able to increase the saturation more with a codec that has more color space to work with before the posterization shows up. This is where transcoding AVCHD footage for example with NeoScene helps out a lot.

Using the video that Camille has posted as an example, there wasn't a whole lot of adjustment made to the chrominance channels other than desaturating them. This is not going to push the limits of any video codec, because you are simply removing information that is already there.

Camille mentioned that the video's contrast was also adjusted. Again, this is not going to break a video codec because every video codec contains 4 out of 4 samples of luminance.

Personally, I would not consider the video by Camille to be heavily color graded, or corrected, or shaded, or whatever you want to call it.
 
I agree with the Chris_TC that there's a lot of FUD going around that material from HDSLRs cannot be graded or colour corrected or timed or whatever you call it. It can but of course like any 4:2:0 material one has to be more careful and know the tools properly.

Here's a example from my commercial shoot with GH1. It was a run'n'gun job and at the spur of the moment we improvised using a outdoor setting. I, dumb as I am, forgot to set the white balance for sunlight in the rush from a indoor set-up, also didn't have time to fix any kind of fill light and for this final take we lost a touch of the sun in a thin cloud so the shadows took a real nose-dive. AVCHD transcoded to PhotoJPEG 8 bit 4:2:2, colour corrected in After Effects with 16 bit colour space and output to a Blackmagic Uncompressed 10-bit 4:2:2 master file. FullHD h.264 Quicktime can be downloaded at Vimeo through my sig-link. More info in this thread: http://www.dvxuser.com/V6/showthread.php?t=186392
 

Attachments

  • sunnycold.jpg
    sunnycold.jpg
    124.4 KB · Views: 0
  • sunnywarm.jpg
    sunnywarm.jpg
    162.7 KB · Views: 0
Great post Chris. Question for you - how did you achieve that look? What steps did you take in terms of color wheels?

:bath:
 
Chris_TC said:
What's your point exactly? Why do you post a clip with a bunch of shallow depth-of-field closeups and crushed whites/blacks?
I picked a landscape scene with tons of micro detail for a reason, but that's apparently lost on you.

The clip you linked to neither has an ungraded version online nor a 1080p version. It contributes nothing at all to this thread.
Click to expand...

okay sorry about my harse comment. I really didn't meant it that way. I thought you didn't liked the quality yourself.
The edges look very harse. That's all i meant.

I still have my first question.

Did you convert the H264 files into a codec with a higher colorspace before grading? ( like ProRes or Uncompressed )
Or did you grade the H264 files directly.

This will affect the final quality of the grade.
 
Last edited:
Some great examples of color grading. I ran into the same issue several weeks ago. Tungsten in the chapel, fluorescent in the reception hall...didn't change my W/B on the 5d2 OR my HVX...looked a mess out of the camera. Final edit, everything matched up near perfectly...both my h264 and DVCproHD:)
I'm very happy with how well it responds. You'd think from most chatter that you'd not be able to touch color in post. Simply. Not. True.

PS...I, too, convert to ProRes 422

Thanks for the examples guys

J
 
im surprised noone has commented about the down rez and uprez - that is VERY promising results - its shows that the original has much more info then the "played with" version - thanks for that part of the test
 
smelni said:
im surprised noone has commented about the down rez and uprez - that is VERY promising results - its shows that the original has much more info then the "played with" version - thanks for that part of the test
Click to expand...

That's a good point. I've read that doing this in post should yield better results than if it were done in camera but here you can clearly see the difference.
 
I downloaded the original file from Vimeo. I'm wondering if you'd be willing to put up the matching 4 seconds of the original video so that we can see how dull it looks and how much noise that has, so that we could see how much is perhaps coming from your grade, and how much is preexisting.

Thanks for putting this together.
 
I've been grading in AE and there is only a very slight visible difference between grading the .mov directly and converting to cineform avi with mpeg streamclip and then grading. That's in a 16 bit project. And the difference is that the avi looks fuzzy compared to the mov with an extreme contrast curve. So it seems better to grade the mov from what I'm seeing. I don't see how having more colourspace in the source footage makes any difference if you are working in a 16 bit AE project, and I can't see any difference either (referring to Paul Jay's statement that grading prores 422 is better). I can't render to prores so I'm unable to compare.

If you drop levels onto the mov vs avi the histogram does look different, but flicking between the two clips at 100% I can't see any visible difference before or after grading between them other than the slight added fuzziness of the cineform conversion. Both the mov and avi histograms look weird in AE. The mov histogram has an evenly spaced missing value every 8 pixels or so, and the avi histogram has evenly spaced downward spikes that don't go to zero.

When I check the histograms in vegas 7, the mov goes 0-255, and the avi is darker and washed out due to the levels only going 14-234. Studio to computer RGB levels effect stretches the levels back to 0-255 on the avi. In vegas neither histogram shows the weird things AE's histogram shows.

Another weird thing with AE (CS3) I have noticed - I set the camera to record in Adobe RGB colourspace, when I bring the mov or avi into AE and set the colourspace to Adobe RGB there is some clipping in the blacks that looks awful. I have to set the colourspoace to none.

I wonder if this problem from the 5D was sorted for the 7D:

http://cineform.blogspot.com/2009/01/full-dynamic-range-video-from-canon-5d.html
 
Last edited:
Thanks for running the test, Chris. Very helpful and answered several questions. It would take an incredible purist to look at this test and conclude the 5D/7D can't be graded.

As Martti candidly pointed out, it becomes a question not so much of camera quality, but how forgiving it is of your mistakes when shooting. As we've seen over and over, it's the carpenter, not the tools.
 
You must log in or register to reply here.
Back
Top