GH4 ALL-Intra -- What is its value?

Michael Erlewine

Well-known member
The PDF GH4 manual describes "All-Intra" as:

"“ALL-Intra” refers to a data compression
method in which each frame is compressed. The file size of motion pictures recorded
in this method becomes larger, but this method can"

I know nothing about this. Is it really better in any way than .MOV? Perhaps someone can weigh in about this who knows please.
 
Hi Michael,

FYI: MOV is a data-container, just like the AVI format is, and it has no innate image qualities all by itself. The video CODEC ( COmpression-DECompression ) that you use to store video images inside the MOV or AVI are what matter.

I've been asking myself the same question with regards to the value of the 1080p ALL-I 200 Mbps footage.

What's the point of having ALL-I 200 Mbps footage, if it doesn't look any better than 100 Mbps IPB compressed footage?

Still trying to figure this out...
 
I was hoping to find someone who had figured it out, of course. I sure don't know. The idea that it can improve resolution intrigues me. Right now i am using 100 Mbps .MOV 4K and it looks pretty good. I am just getting used to this camera. I am amazed at all of the custom adjustment this camera will allow.

I am used to DSLRs (D800E, D7100, etc.), and also BM Pocket, and FS700. If the IQ on video turns out to be reasonable, this is a hell of a bargain for what it seems to be able to do. I also want to get my hands on a Sony A7s if they will ever announce the price and availability.
 
Last edited:
Hi

All Intra means each frame is encoded separately, essentially you get 50/60/24/30 separate jpeg images a second. The other recording modes are all long GOP, which put very simply means one complete frame, then 15, 30 or so frames after are saved as just instructions describing the differences from the first frame, i.e only 1 frame out of a dozen or so is an actual whole frame, the rest are built up from instructions of what has changed.

With all Intra at 60 frames per second which records at 200Mbps then each separate frame is roughtly equivalent to a 1920x1080 JPEG compressed to around 416KB.

Advantages with All-Intra is each frame is pretty consistent in quality. With long GOP, a bit like chinese whispers, frames can vary in quality because the discription of the changes isn't always accurate or complete enough, this is really only noticable though at low bit-rates. Long GOP however can encode the first frame in a group with more bits as the descriptions of the differences after may use very little data depending on the scene, so image quality may be better on some footage and look worse on others.

With All-Intra even if the scene isn't changing very much or is a static image, each frame is encoded as a self contained image, so you waste a lot on duplicated data. This is why recording All-Intra will consistently be around 200Mbps, whereas long GOP is a very varible bit rate depending on the scene.

Editing is much easier on All-Intra because each frame is self-contained, whereas with long GOP if you pick a frame on the time-line the software has to read back to the first whole complete frame, then step through each instruction of frames until it gets to the one required. Modern computers however don't struggle with this too much now.

All-Intra on the GH4 probably may not improve much on the image quality over 100Mbps long GOP unless the footage is very challenging to encode, but with All-Intra you sort of get a guaranteed quality, will not be better or worse, whereas long GOP may be worse on some scenes, but then again may be better on simple non-complex scenes.

Regards

Phil
 
All-Intra on the GH4 probably may not improve much on the image quality over 100Mbps long GOP unless the footage is very challenging to encode, but with All-Intra you sort of get a guaranteed quality, will not be better or worse, whereas long GOP may be worse on some scenes, but then again may be better on simple non-complex scenes.

Yes. For example, when panning, every frame is completely different from every other frame, which is sorta the definition of panning (or tilting, or dolly moves). Or if the camera isn't moving, but everything in the frame is, like moving water, or the wind blowing leaves in a tree. These kinds of footage, are why Intra-frame CODECs exist.

Where LongGOP CODECs excell is when the camera isn't moving, and some of what's in frame isn't moving. For example, a talking head on TV. Anything on a set where only the talent is moving. For these kinds of footage, a LongGOP CODEC saves a lot of space. And the savings can be huge.

Typically, a LongGOP CODEC is thought of as a capture CODEC. In the not-so-distant past, people would transcode a LongGOP CODEC into an Intra-frame CODEC for editing, because it makes editing much easier (as in, takes less computer horsepower to accomplish). But in the last few years, this has become nearly a moot point; transcoding has become less important to a wider range of NLE workflows as the NLEs have become better at handling LongGOP CODECs.
 
What great answers! If I understand correctly, then ALL-Intra is not variable in size, but fixed. That being given, we can expect the recording time/storage for ALL-Intra to always be the same. I see why we would want to use it for a rushing waterfall since almost everything in the frame is changing each frame. My question then becomes why does the GH4 not offer All-Intra at 100 Mbps, but only at 200 Mbps?
 
In theory all-I should look better with motion given the frames have enough data rate which 200mb/s should do.

Yes, I know the theory, but I have yet to see footage that puts this theory into practice. It would be useful to know if there are any practical shooting conditions where this 200 Mbps will actually show a noticeable difference.
 
then ALL-Intra is not variable in size, but fixed

That's the theory, but from what I've read online, some people are still seeing small data-rate differences when shooting GH4 200 Mbps ALL-I footage. I suspect these difference may stem from not enough detail in the subject they are shooting. ( i.e. What's the actual recorded data-rate when shooting a blank evenly lit white wall ? )
 
It would seem that that could almost be one way of measuring camera noise?

It's usually darker tones and blacks where noise becomes an issue, which is why proper lighting is key to getting noise-free shots.

I wonder how often ALL-Intra forces a full frame if there is no further variation?

In theory every frame should use the same amount of storage, so ALL-I should not vary it's data-rate.
 
It is inaccurate to say all-I has the same level of quality for every frame. It has the same bitrate for each frame but sometimes a frame may need more or less. Just like with jpeg image compression difference types of images compress at different rates. For example a jpeg of a blue sky may be 100kbps while a jpeg of a forest the exact same resolution could be 500kbps or higher.

All-I can suffer from bitrate starving and compression artifacts just like IPB footage. Just different type of artifacts. This is why the 72mbit mode on the GH3 isn't all that great. It is just too low of a bitrate to handle the image quality capable with that camera. All-I can suffer from noise along edges when the bitrate is too low. This is known as mosquito noise and was very common on formats like DV which was an all-I format as well.

With that said 200mbits is a lot for HD and should for the most part result is overall great quality depending on the situation. You may get a bit more noise in insane detail scenes but for the most part should give you consistent quality. This is because most of your frames need nowhere near 400kbps per frame.

The reason why IPB even at 50mbps can look just as good is because the bits are shape red between a group of frames and motion predication is used to help share blocks of pixels. In that case where a blue sky frame may only need 100kbps instead of 500kbps the IPB version would only use enough bits as are needed. All-I will throw 400kbps at every frame even if it is a solid black frame recorded with the lens cap on. So IPB is more efficient because it tries to figure out how many bits each frame should use. This is why for 90% of shots in a typical movie IPB at a high enough bitrate can look virtually perfect. Where IPB starts to fall apart is if that bitrate limit isn't high enough for complex motion scenes. Scenes where the motion prediction cannot keep up with the 50mbos it has to work with. Then it starts to just thrown whole chunks of pixels instead of calculating each pixel. This is when you see macro blocking. The higher the IPB rate the more it eliminates the need for all-I. 50mbps is extremely clean. 100mbps IPB is insane clean. Anything over that for IPB is a waste because the motion predication ends up doing the same thing as the all-I format would do. It has so many bits to work with it just throws the full amount at the image anyway.
 
If I understand correctly, then ALL-Intra is not variable in size, but fixed.
No, you've gotten the wrong impression. The only things that are fixed about All-Intra recording are the frame rate and the maximum bitrate of the video stream. The bitrate is not divided equally between frames, it is monitored and regulated over time. Individual frames are compressed at the highest quality that can be sustained at the overall bitrate, and this will vary according to how detailed, brightly lit, and sharply focused the scene is. As details change over time, the amount of data used to encode each frame may change as well, and it typically does not require the full 200Mbps to encode most scenes.

While long-GOP IPB encoding formats can potentially deliver comparable frame-to-frame quality as All-Intra, consumer-grade codecs are optimized for minimum file size rather than maximum image quality. They will typically encode P-frames at a lower quality than I-frames, and B-frames at a lower quality than P-frames. This is based on the assumption that since P and B-frames contain only a fraction of the details encoded in an I-frame, their quality can be lowered without noticeable image degradation. I think this is an acceptable compromise for amateur videos that are intended for viewing as-is, but is inadequate for capturing video images that will be professionally edited or enlarged.

Tests on the hacked GH2 showed that 1080p All-Intra recording required at least 150Mbps to maintain I-frame image quality as high as the I-frames in long-GOP formats. In my view, Panasonic made a wise choice allocating up to 200Mbps for the GH4's All-Intra codec. The GH2 was not able to reliably sustain such a high bitrate, and All-Intra patches were notoriously difficult to optimize for consistent image quality under all circumstances. With Flow Motion v2, I concluded that 100Mbps was a more practical bitrate for the GH2, and that ruled out an All-Intra format. What I used instead was a short-GOP format only three frames long, each GOP containing an I-frame followed by two B-frames. I then manipulated the codec's quantization tables to force it to record the B-frames at comparable quality levels as the I-frames. That, combined with the fast motion-tracking attributes of the short-GOP format, produced a compact, high-quality solution well-suited to the GH2. With the larger and faster SD cards available today, the 200Mbps All-Intra format is a great choice for the GH4.
 
Would I be right in assuming that most delivery formats will be long-GOP, such as vimeo, tv or a blue ray? Or can it be All-Intra as well?
One thing I would like to know is when an editor program is exporting a long-GOP video does it 'respect' a long-GOP input file and use I frames for I frames or does it just reassign I frames to whatever frame happens to fall at a particular point?
If it is the latter, then All-I capture would make sense for this reason because you don't get progressive degradation of the footage.
 
Yes, long-GOP H.264 was developed as a commercial delivery format, e.g. Blu-ray video, where file size and bitrate streaming limitations override image quality concerns. In those cases, look-ahead techniques can be used to anticipate frames with high bitrate demands and optimize bitrate usage accordingly. For real-time recording, however, a long-GOP format is poorly equipped to handle sudden scene changes that may occur in the middle of a GOP. That is the primary advantage of All-Intra and short-GOP formats for video capture.

When you import a compressed video into an editor, it's decoded into a standard internal format independent of its original format. When you render the video in an export, it's freshly encoded from the editor's internal format, without regard for the original format. In cases where it's important to preserve as much of the original data as possible, specialized transcoders can be used to convert directly between related codecs, but such transcoders are not commonly built into general purpose video editors.
 
Hi

Yes long GOP as suggested is really a delivery format, and now also used as an acquisition format, mainly because the LSIs are fast enough to make a good job of encoding the more complicated long GOP format in real-time and it reduces bit-rates while keeping the quality expected so that consumer SD cards can be used.

All-Intra is really targeted for acquisition in professional equipment where high speed memory formats come into play to support high bit-rates (although now possible on more standard SD cards), for example Panasonic's AVC-Intra format is, under the hood, H264 All Intra.

Interestingly enough AVC-Intra 100, Panasonic's professional recording format, is only 100Mbps recording and supports 4.2.2 and 10 bit colour, yet on the GH4 we have 200Mbps but only 4.2.0 at 8 bit. We might suppose that 4.2.2 at 10bit is perfectly possible in HD to be recorded internally on the GH4, but Panasonic didn't want to make it that good in comparision to their much more expensive pro-gear. Maybe a hacked firmware will enable this later.

Regards

Phil
 
Panasonic's professional recording format, is only 100Mbps recording and supports 4.2.2 and 10 bit colour

Panasonic also have AVC-LongG50/25 which supports full raster HD 10-bit 4:2:2 format at data-rates as low as 25 Mbps, but I think this requires a higher level of processing power than the GH4 currently offers. I'm sure it's possible to produce a GHx camera that supports this CODEC, but are people willing to pay a significant premium for this format ? ( i.e. Double the price of the current GH4 )
 
Hi

Panasonic also have AVC-LongG50/25 which supports full raster HD 10-bit 4:2:2 format at data-rates as low as 25 Mbps, but I think this requires a higher level of processing power than the GH4 currently offers. I'm sure it's possible to produce a GHx camera that supports this CODEC, but are people willing to pay a significant premium for this format ? ( i.e. Double the price of the current GH4 )

I don't think it would cost Panasonic twice the amount of course, I'd not be surprised if the LSI in the GH4 is the same LSI used in their professional cameras supporting those formats, but of course Panasonic would want to charge much more to include these as otherwise it would impact sales of their higher end cameras.

But if it is possible to modify the firmware to support this unofficially, I'm sure someone will. :cheesy:

Regards

Phil
 
Panasonic also have AVC-LongG50/25 which supports full raster HD 10-bit 4:2:2 format at data-rates as low as 25 Mbps, but I think this requires a higher level of processing power than the GH4 currently offers. I'm sure it's possible to produce a GHx camera that supports this CODEC, but are people willing to pay a significant premium for this format ? ( i.e. Double the price of the current GH4 )

I would :)

Heck, there are already plenty of threads in the AF100 section and others asking for just such a camera (GH4 front end on PX270 back end).
 
Back
Top