EVA1: Differences between Long GOP and All I

[Noel Evans]

Im going to drop some comparison footage soon, but I must say seeing the ALL I shows how good the Long GOP actually is. Yes, if I pixel peep I can see a difference and in some cases more than others (will show with footage), but seriously if you never had an actual need / demanded by production you could get away with Long GOP all day long. That said Im happy to have the ALL I as I do have a need.

YMMV, any others have thoughts on this?

Will drop clip in here, as soon as I get a chance to do export and upload.

UPDATE: Not sure what purpose this now serves as its compressed but here are the clips I shot in my comparison.

 

[Cary Knoop]

Of course, there is nothing wrong quality wise with inter-frame compression!

There seems to be so much misunderstanding about compression, from bad to worse:

1. H.264 is a bad quality codec.
2. Long GOP makes your footage look bad with many macroblocks.
3. Intra-frame only encoding gives a better quality than a combination of intra-frame and inter-frame encoding.
4. Inter-frame compression messes up your cadences.
99. ProRes is the best, it makes your footage Oscar Ready™ by gold coating your footage, all other codecs fail in comparison, and it is not subject to discussion. Apple is the best, yeah!

The most efficient compression is a combination of intra-frame and inter-frame compression.


The only disadvantage with inter-frame compression is that the decoding may require multiple frames which is not a problem when playing the footage but it is a problem when you are seeking, adding or removing frames which happens when you edit a video.

 

[bill totolo]

Noel, can you push the Intra frame codec further in post before it breaks?
Is it better under a limited spectrum light source such as sodium vapor?

 

[Seanik]

99. ProRes is the best, it makes your footage Oscar Ready™ by gold coating your footage, all other codecs fail in comparison, and it is not subject to discussion. Apple is the best, yeah!

The only downside is that it farts rainbows and glitter! ...eh, I'm more than happy to avoid having to use ProRes if I don't have to.

 

[Thomas Smet]

Yes there is a ton of misunderstanding out there thanks to the inaccurate blogs and YouTube videos that exponentially spread the misinformation.

The biggest thing that drives me nuts is when people compare IPB to all-I with a locked down camera shooting a still life scene. IPB formats save data with motion prediction with the key word being "motion". A still life scene will of course look stunning at 150mbps and 400mbps and neither format is going to really look better. Where the 400mbps "may" shine is when you have insane random motion like a large fiery explosion or a hurricane environment. Codecs are so good today that even in extreme situations it is difficult to break them. It is almost impossible to even compare a side by side difference in the kind of situations that break either codec. With that said there may be some situations where the 400mbps holds up better but it may be splitting hairs once you get to that point since it would take a really jacked up environment not optimal for shooting anyway.

My point is it is going to take one hell of a stress test to really compare the difference between the two. Not sure anybody is ever going to be able to test that sort of thing accurately unless they happen to be shooting during the next hurricane and take time to think of grabbing some 150mbps and 400mbps shots vs saving their lives. But hey if they do and can manage to upload the results before they die my thanks to them.

Seriously the only way we will ever see a real difference if any is to create a very complex fast, sudden, random motion test that has the impossible task of breaking the codec. Even 400mbps can break in some situations, just I na different way. Instead of macro blocking you get edge mosquito noise and a DCT form of macro blocks. Even jpeg still photos have macro blocks if the data is too low and the same can happen for all-I. Thats why there is a ProRes LT, ProRes and ProRes HQ flavor. Different quality levels of the same all-I, 10bit 4:2:2 style of compression. 400mbps is roughly the data rate of ProRes LT yet appears to look much better than LT does in high detail scenes like grass and foliage. I'm not entirely sure I would call it as solid as ProRes HQ but then again I have not been able to shoot in a hurricane or blow up my house to test it out.

 

[Barry_Green]

In general, the 400mbps All-I should deliver picture quality comparable to the 150mbps LongGOP. The substantial differences between the two are going to be in file sizes and in editing performance.

Lots of people despise long-gop because it was first employed in HDV, the worst recording format ever introduced to mankind. It was a case of trying to push it way, way, way farther than it ever should have been pushed, all in an attempt to make high-def footage fit on a miniDV tape. It was asking too much, and the long-gop of HDV couldn't keep up, and the footage broke in a lot of reasonably common situations. As such the whole notion of long-gop has been spoiled for a lot of folks.

It doesn't need to be that way. The concept of LongGOP is fine and solid, it just needs to be implemented appropriately. With adequate bandwidth, LongGOP can be a very effective compression technology.

Generally, LongGOP can be more efficient compression to the point where ALL-I requires at least 2x to 3x the bandwidth to match the efficiency of LongGOP. At 400 mbps, the ALL-I is 2.7x the bandwidth of the LongGOP, so -- it is going to compare very favorably indeed.

The 200mbps ALL-I is only 2x the bandwidth of the 100mbps LongGOP, so ... that may be a case where the LongGOP may actually still look a little better.

But the All-I should be much faster to edit. That is the main benefit to it.

 

[Seanik]

LongGOP h.264 was almost guaranteed to break on me in FCP7, CS6, and fairly often in the first couple of versions of CC. Those are relatively recent bad experiences. It's not so bad anymore, but I still dread having to edit with it (on a 12 core MacPro). Premiere's proxy workflow is (4K) LongGOP's only saviour. That said, it's still just a workaround for a compression scheme that was designed for delivery, not acquisition.

 

[Noel Evans]

Ugh I had done a long long post that I lost...

Here's the abridged version.

I shot a bunch of shots side by side - All I and followed with Long GOP. No setting changes whatsoever. On an overcast day this meant some slight exposure variances between the two, however I didn't want to adjust exposure as it would muddy the results. I shot a bunch of daylight shots in settings that would be a little noisier than usual. I then added the rec.709 LUT and did comparisons. I then did some crazy grading in resolve to really push it. I then did some normal grading in resolve. Rinse and repeat in FCPX. In every way, they are quite comparable. I didn't see any great benefit of one over the other. I tried.

Personally I like working with the ALL I and my machine prefers it as well. Fast HDD space is cheap these days.

OK so heres a clear example of where ALL I was better, some may laugh at this due to how far in I had to go. Look at the image below. The enlarged extracts are of a yellow barrier. The barrier had diagonal lines that converge. There's a few of them in the image and they are quite a distance from where I was shooting. They are tiny in my image to start with. If you look at the all I you can see the diagonal lines all stay nice and straight in the centre, however on the long gop, there's a few muddy spots. Pretty easy to see what I am talking about. And this is the variance I found. If you look at the buildings in the BG, on the all I image these too were more defined. To reiterate, I did not change any camera settings at all.

Had to make 2k version to fit here which means you can barely see the red square where the extraction is from, you can still see the variance Im talking about...

The extractions are 300% of original.