Ad
Ad
Ad
Pages: [1]   Bottom of Page
Print
Author Topic: more is less ?  (Read 3438 times)
fredjeang
Guest
« on: May 26, 2012, 02:09:01 PM »
ReplyReply

Hell and damnation !

The more I progress, the more I work, the more I think I know, the more I realise I don't know.

How can it be ?

Again in the doubt and confusion in things I beleive were settled.

After reading several times in several different sources experienced people having done some testings with the different version of Prores.
Sorry, Talking about the on-line editing here. Color and finishing.

So what's the prob? Well according to their testings, the Prores HQ increments drastically the file size but nobody can really tell the difference on screen.
Between the 444 and the 422 same thing.

In fact, same with Avid between the different versions of DNxHD. There is a moment where I stop to perceive an increment quality. And the doubt is that if those ultimate codecs are here it is for a reason I guess.
Where does really start and stop the increment in quality? I'm of course not talking about the proxies but the finishing.

I did also testings in the recent past with DNxHD 10 and 8 bits and HQ-HQX and really, really can't tell the differences, even in grading. I don't see it.
Maybe my eyes are failing but I don't perceive any real increment while yes the file size increments a lot.

So my question is: What's the point to use the highest codecs? Is the increment in quality really there? Isn't it better to work in 4:2:2 instead?
Same doubt with those 8-10 bits.


In other words, in what to grade-finish keeping in mind the best quality-file size? I mean not compromising the quality.

Not the biggest seems to be necesarly the best.

What are your thoughts?


 
« Last Edit: May 26, 2012, 02:16:43 PM by fredjeang » Logged
fredjeang
Guest
« Reply #1 on: May 26, 2012, 02:27:52 PM »
ReplyReply

To use a cheap analogy:

You are in a supermarket where there are many different sort of fishes (codecs).

The dealer says: get this discusting low-end fish at 1 and with it you preapare your meal (off-line)

Once the meal is ready, you'll buy some more expensive fishes and redo (EDL) your meal with them so it tastes good (on-line).

This is where the problem is: on the most expensive fishes, some cost 5 , others 15 (the cost is the file size)
The 15 are supposed to have much better taste than the 5's...no? aren't they more expensive ? but when you taste then, you hardly notice the difference.

The doubt comes because you think, well, I don't notice the difference but maybe my taste is failing. Being more expensive, it should taste better.

With what to cook then ?

And more...is there a fish with wich you can both preapare easily (edit) and taste good (finishing) ?

The goal is:

- doing the less possible transcoding
- a suitable format for both editing and finishing (no proxies needed). Here I found that the Canopus HQ is the one that worked best for me.
- maintaining the image quality at the max for grading and finishing
- while maintaining the file size at the smallest possible
- and that doesn't loose too much when doing families (think of jpegs here. Each time you write a jpeg it degrades)

As you see, I'd also like to avoid the re-linking to the high res files, but I'm aware it's not necesarly the best option to keep the qualy at its highest.
« Last Edit: May 26, 2012, 02:52:23 PM by fredjeang » Logged
Chris Sanderson
Administrator
Sr. Member
*****
Offline Offline

Posts: 1906



« Reply #2 on: May 26, 2012, 05:07:20 PM »
ReplyReply

To extend your analogy, the cheap fish will fall apart to mush quicker than the more expensive stuff.

Do some extreme grading in 444 and you will likely see less quality loss than in 422 or lower. Otherwise it's all the same paella.

Post grading? I can't see much difference on a high-end pro monitor, on a regular good monitor - none at all.
Logged

Christopher Sanderson
The Luminous-Landscape
smthopr
Full Member
***
Offline Offline

Posts: 237


WWW
« Reply #3 on: May 26, 2012, 08:08:49 PM »
ReplyReply

I think you'll notice the difference between 444 and 422 most when pulling mattes for green screen compositing. Having the extra color detail can really help.

For other stuff the difference is much more subtle, and probably not worth worrying about.

That said, I did once notice a visible difference between 444 hdcamSR and 422 prores of the same material. The prores was more "dirty" looking to my eyes.

I also graded some mixed material and when the 8 bit stuff comes on compared to the 10 bit, it's noticeable when significant grades are made. Kind of a subtle change in tonality.

With all that said, the 444 , 10bit original graded, then compressed to blue ray looks really good on the highly compresses blue ray.  If I played out the original 444 10 bit and then the blue ray, I doubt anyone would notice the difference.

So if you don't find yourself needing the uncompressed 444, save your money and bandwidth until you need it, I would say, and as you have observed.
Logged

Bruce Alan Greene
www.brucealangreene.com
fredjeang
Guest
« Reply #4 on: May 27, 2012, 04:21:22 AM »
ReplyReply

Chris, Bruce, many thanks.

The informations you provide match my experience.

I can see a difference using 444, but basically in greenscreens for ex, things like that.
But do not notice any difference between Prores 422 and 422 HQ.

It happens exactly the same way with the Canopus codecs. Did over and over again testings with the HQ and HQX, nothing ! not even in subtle transitions, not even grading the brute way. But yes the use of a 10bits codec that expands the range has an impact, not positive, on the editing.

Oh well, the increment in file size between HQ / HQX ; Prores 422 / 422 HQ is not that big, a 30ish % so one or the other really doesn't matter in terms of storage. Yes it matter in how they stress the softwares.

444 doubled the size of 422. I guess the best bet is to reserve 444 for theaters because even there the gain-stability is marginal.

I also smell that those codecs with .264 files of the 5Ds and GH2s aren't really doing miracles. I mean that the 8 bits highly compressed footage is what is and almost nothing can be expected. For example, I did some testings with the banding (wall scene with smooth gradient). No matters what codec you use with a Canon or Pana, the moment when the image falls appart is the same. No extra room for playing.
I'm not saying that what I just pointed is true, but I can tell you that I'm absolutly unable to notice anything.
With 444 yes I do notice it stands better, but it's overkill in most of the cases.

So, are you close to think that using Prores 422, not the HQ, is the best 4x4 in the sense that it suits both the editing and the finishing? That's what I'm not far to think.



 
Logged
smthopr
Full Member
***
Offline Offline

Posts: 237


WWW
« Reply #5 on: May 27, 2012, 12:17:53 PM »
ReplyReply

Fred, what format are your camera original files?
Logged

Bruce Alan Greene
www.brucealangreene.com
fredjeang
Guest
« Reply #6 on: May 27, 2012, 04:06:47 PM »
ReplyReply

Fred, what format are your camera original files?

Generally MTS at 100 MB/s, (I've been shooting MTS at almost 200MB/s and it's completly meaningless with such a compressed file, unless you're shooting B&W at higher isos you'd have a slightly more pleasant grain)

Sometimes, less frequent, AVC-Intra (the 100 version).

I've also been doing those testings on P2 footage one time, same conclusion.

The only format where I found a huge difference and plenty of room when it comes to post is the Red.


PS: In my previous posts, I was talking strictly about the on-line editing. Evidently in adquisition there is an enormous gap between the .MTS and the Intra-100, despite both being recording at a max of 100 MB/s.
So yes the original file of the Intra is clearly superior and clearly visible on smooth gradients. But then the observations I made in post on the codecs match in the same sense for bvoth. The only difference being the fact that the Intra falls appart later (and not that much later, it's not a R3D).

I think that both Red and the Alexa are really the very best devices. The Alexa being more straightforward. But I'm not really an expert with those 2 cameras because if I've edited on them, I don't own one and therefore lack of hours.

My best bet for 90% of the cases is using ProRes 422 but not the Hq or Canopus HQ and not the HQX.  In DNxHD I have mixed feelings and need to test more to find the best compromise.

DNxHD interests me because I want to use it in Lightworks when they'll release the commercial version those days, as I'm sort of familiar with this codec in Avid.
« Last Edit: May 27, 2012, 04:19:13 PM by fredjeang » Logged
smthopr
Full Member
***
Offline Offline

Posts: 237


WWW
« Reply #7 on: May 27, 2012, 04:22:34 PM »
ReplyReply

Fred, please excuse my ignorance. What is MTS?

From the p2 comment, I assume you're shooting with a Panasonic camera.

I guess what I'm getting at is are you shooting 10 bit 422?
Logged

Bruce Alan Greene
www.brucealangreene.com
fredjeang
Guest
« Reply #8 on: May 27, 2012, 04:25:33 PM »
ReplyReply

Fred, please excuse my ignorance. What is MTS?

From the p2 comment, I assume you're shooting with a Panasonic camera.

I guess what I'm getting at is are you shooting 10 bit 422?

MTS is AVCHD extention. A plague !

Yes it's a Panasonic too.

Yes it's 10 bits 422. Correct. (the Varicam That I used sometimes in some projects not the GH2 (my personal camera) wich is AVCHD 8 bits)


There is something in all that I notice. What really changes is when I shoot 10bits vs 8 bits. There yes there is a difference. But if those 10 bits are added in post, knowing that they don't add any info but I expected that the grading would be more pushable, I can't see it.
In other words, it's on the recording stage that all happens. At least it's what my own ignorance tells me.
« Last Edit: May 27, 2012, 04:38:24 PM by fredjeang » Logged
Morgan_Moore
Sr. Member
****
Offline Offline

Posts: 2218


WWW
« Reply #9 on: May 28, 2012, 08:18:42 AM »
ReplyReply

It all seems completely sensible but those who say they cannot tell the difference are doing the wrong tests

The test should be to shoot a scene with 3200 lighting and the camera set on 5500

Correct that and you see the difference!

Being serious we know there are significant challenges to monitoring 'on scene' unless you have a cart with a proper monitor, and a black tent and three guys to move it

I would personally be seeking enough flexibility in a file that if it looked good on set then one can make it look good on screen

S
Logged

Sam Morgan Moore Cornwall
www.sammorganmoore.com -photography
smthopr
Full Member
***
Offline Offline

Posts: 237


WWW
« Reply #10 on: May 28, 2012, 01:19:48 PM »
ReplyReply

If you're shooting 10 bit, I would for sure use a 10 bit codec such as ProRes422hq.

As for correcting wb like shooting tungsten with daylight balance, then 10 bit recording will not be enough to save you. You'll need to shoot with enough dynamic range to avoid clipping any color of your highlights. You could do this on the varicam if you shot very low contrast using dynamic level of 400% or greater. but when you shoot high dynamic range, you'll really want to keep everything 10 bit until the final output

With the GH2, you're probably out of range for a full correction.
Logged

Bruce Alan Greene
www.brucealangreene.com
fredjeang
Guest
« Reply #11 on: May 28, 2012, 01:35:45 PM »
ReplyReply


With the GH2, you're probably out of range for a full correction.

Absolutly Bruce. I found the same with the 5D2.
Between a 5D2 and a GH2 I choose the Panasonic. But there are serious limitations as you point. (well, after all those are consumer cameras, can't complain)
The Varicam (I don't own !) isn't my cup of tea I must say. It's a superb camera but not for me, very broadcast orientated, big and heavy and overkill for the tecnical level I am so far. Still very clumpsy with those Film-REC etc... Here in Madrid you find them complete with Fujinon zooms at more or less 30.000 second-hand. It's money but not too much considering the level. I did few 720 projects and IQ is stunning and you get 14bits in processing, but I really prefer the Reds-Alexas philosophy. I'd like to shoot directly Prores 444 like the Alexa does and end of the hassles. Red of course tempts me but There are steps in the workflow. I'd like a simplified-straight workflow at the max, preferably all on-line.

I still have a big doubt with the new interesting Sony that looks ideal on the paper considering price-performance but it features also 8 bits . Footages look good indeed but ...

And thanks for the info that keeping 10 bits on HDR to the end and not downgrade codec to gain flexibility. That makes sense. It was helpfull.


To resume, please correct me if I'm wrong.

10 bits in adquisition - 10 bits workflow in post
8 bits in adquisition- 8 bits workflow in post, 10 bits added in intermediate codec won't change anything and image will fall apart the same way. Unless alpha is needed 10 bits on-line from a 8 bits material is pointless.

Correct ?


Ps: a colorfull post.
« Last Edit: May 28, 2012, 02:51:42 PM by fredjeang » Logged
bcooter
Guest
« Reply #12 on: May 28, 2012, 03:40:58 PM »
ReplyReply


snip........

To resume, please correct me if I'm wrong.

10 bits in adquisition - 10 bits workflow in post
8 bits in adquisition- 8 bits workflow in post, 10 bits added in intermediate codec won't change anything and image will fall apart the same way. Unless alpha is needed 10 bits on-line from a 8 bits material is pointless.

Correct ?


Ps: a colorfull post.

Not in the real world.

A few years ago we had some "challanged footage" from a shoot in Jamica featuring a sports celeb.

Light changed, the camera operator shot it under, the subjects were dark complexion and all I had were silouettes.

I tried to open it up in FCP with filters, Apple Color, in an Avid System, all with the same results.   Track noise, blown out BG, etc.  

Then finally went to a Di-Vinci suite (the old one the size of a large room, where a top rate motion picture colorists put the tapes (captured in 8 bit) into the 12 bit di-vinci and she opened it up, colored it, corrected it and tracked it with no problem, in fact it was almost pretty, which was a heck of a leap considering it was non usable before.

So to be honest at that point I didn't care if it was faux 12 bit, real 12 bit, I just cared that it worked which it did, I got paid, life went on.

IMO

BC

P.S.    On the session mentioned the operator I used was skilled and had a great resume, but back then High Def was limited to a few cameras.  We used a rented Canon XL series and a PS technique adapter and a rented clamshell monitor.  When he shot, something looked funky in the evf and the monitor and I yelled cut.  I went back played the tapes and just didn't trust what I was seeing.  I told the operator/dp to yank the high def, put my trusted xl1 on the dolly and shoot with it, because I knew the evf was dead on with that camera.  He resisted, he held his ground, he told me it was fine, but in my gut I knew it was off.

That not trust my gut cost me 7 grand in correction and I learned a huge lesson.  

#1 either own your own equipment or test everything first (in a computer).

#2  Always trust your gut.

#3  Don't let a group of 15 clients and a difficult subject hurry you into a mistake.
Logged
fredjeang
Guest
« Reply #13 on: May 28, 2012, 04:47:19 PM »
ReplyReply

Interesting story that ads a valuable contrast in the thread.

It's true that I've never did any comparaison in such an extreme scenario.
In general so far, in practise, I've always been using the max quality codecs available.

As you point, I couldn't care less either if bits are real or false, in the moment they really give more room and-or can save situations, they are usefull.

So I'll be doing testings with just the same scenario you mention because if I could not detect (and not only me, this is an old topic in many video forums) any magical increment in normal situation,
where footage was equilibrate, maybe in delicate situations as you point and that can happen to everybody,  there may be a difference.
But my little finger (or gut to use your expression) is somewhere telling me the results in advance.

It still surprises me this Da-Vinci stuff, because if the information recorded isn't there it isn't there, 8 or 500 bits. I'm not sure if in fact the information that the woman recuperated was in fact recorded but for some reasons it could not be recovered but just in Da-Vinci.

Is it really because of the 12bits boost coming from a 8 bits original file?  That's precisely the doubt and the reason of this thread.

It could be that Da-Vinci was a simply more advanced software capable of extracting until the ultimate of your 8 bits footage and that its engine was more elaborate.
If she could save the footage, it's because it was potentially there already and recorded. Da-Vinci didn't-couldn't invent anything adding some bits.

Weired but interesting.


Ps: I've been scanning the specialists forums on that topic, trying to extract the posts from the most experienced old foxes of this industry and there is really no consens. The general idea resumed is more or less in alignement to the apportations of the posters in this thread but the motion pros are far to agree between themselves on this subject.
 
« Last Edit: May 28, 2012, 06:26:59 PM by fredjeang » Logged
bcooter
Guest
« Reply #14 on: May 29, 2012, 11:17:58 AM »
ReplyReply

I don't think my results had much to do with 8, 10 or 12 bit.

I think the real difference was this is the results when you use the million dollar "di-vinci is a room suite", the colorists works on academy award winning production and has 25 years experience.

The interesting thing that is when we opened the footage up ourselves, using different suites, different plug ins there was either a lot of track and snow like noise or using plug in's too smooth noise killer that lost detail.

The colorists could easily get to the right mix and deliver a really nice result.

Nothing beats experience and a suite, dedicated to only one function.

IMO

BC
Logged
Pages: [1]   Top of Page
Print
Jump to:  

Ad
Ad
Ad