GH5 Evaluating GH5 4K 10bit clips recorded at both 150Mbps and 400Mbps.

joe12south said:
2. It removes the chance of synthetic motion artifacts. Long-i codecs predict motion, and sometime predict incorrectly in ways that are visible. (On the GH-series, such an artifact can be induced by panning, for example.)
Click to expand...
The motion between two frames is basically estimated, then the frame is constructed based on this estimated motion, after this it is compared with the actual frame then only the difference is compressed.
A bad motion estimation will not incur more artifacts by itself but it will make the compression of the difference between the frames less efficient (which could incur more artifacts).
 
Cary Knoop said:
The motion between two frames is basically estimated, then the frame is constructed based on this estimated motion, after this it is compared with the actual frame then only the difference is compressed.
A bad motion estimation will not incur more artifacts by itself but it will make the compression of the difference between the frames less efficient (which could incur more artifacts).
Click to expand...
Here's the catch: P-frames are constructed from blocks found in previous I or P-frames in the same GOP. These blocks are taken not from the original image details, but from the reconstructed I and P-frames produced by the decoder. (H.264 encoders contain decoders as well, used to reconstruct frames for this purpose.) As stated above, discrepancies between predicted blocks and original image details are then intra-encoded into each P-frame. Consequently, in long-GOP encoders, artifacts encoded into I and P-frames can be propagated into the predicted blocks used in subsequent P-frames. These errors can accumulate in a daisy-chain manner from P-frame to P-frame. Additional bitrate is then required to correct these extraneous artifacts with intra-encoded data. In frames that contain a large amount of detail and/or motion, the transient bitrate budget may be exceeded, and traces of these artifacts may become perceptible (AKA "macroblocking").

In All-I encoders, artifacts are guaranteed to be limited to intra-frame encoding discrepancies, and are never propagated into subsequent frame predictions. I made use of this principle in my GH2 Flow Motion patch, which is a short-GOP encoder composed only of a single I-frame followed by two B-frames in each GOP. The predictions used in each B-frame are from the two surrounding I-frames, limiting prediction artifacts to a single generation of intra-encoded blocks, at most two frames distant from each B-frame.
 
Last edited:
Lpowell said:
Here's the catch: P-frames are constructed from blocks found in previous I or P-frames in the same GOP. These blocks are taken not from the original image details, but from the reconstructed I and P-frames produced by the decoder. (H.264 encoders contain decoders as well, used to reconstruct frames for this purpose.) As stated above, discrepancies between predicted blocks and original image details are then intra-encoded into each P-frame. Consequently, in long-GOP encoders, artifacts encoded into I and P-frames can be propagated into the predicted blocks used in subsequent P-frames. These errors can accumulate in a daisy-chain manner from P-frame to P-frame. Additional bitrate is then required to correct these extraneous artifacts with intra-encoded data. In frames that contain a large amount of detail and/or motion, the transient bitrate budget may be exceeded, and traces of these artifacts may become perceptible (AKA "macroblocking").

In All-I encoders, artifacts are guaranteed to be limited to intra-frame encoding discrepancies, and are never propagated into subsequent frame predictions. I made use of this principle in my GH2 Flow Motion patch, which is a short-GOP encoder composed only of a single I-frame followed by two B-frames in each GOP. The predictions used in each B-frame are from the two surrounding I-frames, limiting prediction artifacts to a single generation of intra-encoded blocks, at most two frames distant from each B-frame.
Click to expand...
True!
 
Is there any good comparison GH5 all-i vs ipb? In my mind the ipb wins in video quality because it compresses less the i-frames. I have shooted thousands of 4k 60P videos with high shutter speed and I cant see any disturbing faults when watching stopped frames. Every frame is like a high quality photo.
 
Vesku said:
Is there any good comparison GH5 all-i vs ipb? In my mind the ipb wins in video quality because it compresses less the i-frames. I have shooted thousands of 4k 60P videos with high shutter speed and I cant see any disturbing faults when watching stopped frames. Every frame is like a high quality photo.
Click to expand...

That's been my question - even though the bitrate is higher with the ALL-I codec it has to compress every frame whereas IPB is only applying compression to selected frames so the compression on these may be less?. What I need to do is to shoot a difficult tone like a finely graduated clear sky in V-LOG with both codecs and then see which one breaks first when heavy grading is applied......
 
GH3/GH4 200Mbs all-i 1080P was worse than 50Mbs ipb. How can 400 Mbs all-i be good when there is 4 times more pixel data than in 200Mbs 1080P all-i ?
 
Vesku said:
GH3/GH4 200Mbs all-i 1080P was worse than 50Mbs ipb. How can 400 Mbs all-i be good when there is 4 times more pixel data than in 200Mbs 1080P all-i ?
Click to expand...

Not saying it is better, but don't fall into the trap of assuming that data rate required to reach PQ equivalency scales linearly. It most definitely does not with any codec I know of in current popular use.
 
Shirozina said:
That's been my question - even though the bitrate is higher with the ALL-I codec it has to compress every frame whereas IPB is only applying compression to selected frames so the compression on these may be less?. What I need to do is to shoot a difficult tone like a finely graduated clear sky in V-LOG with both codecs and then see which one breaks first when heavy grading is applied......
Click to expand...
You're not going to find a meaningful PQ difference. Panasonic didn't choose those bitrates arbitrarily. They are both proportionately sufficient for the majority of scenes.

Choose all-i in order to layer more realtime 4K effects, eliminate motion artifacts or because you want to give more money to Seagate or Western Digital. ;-)
 
I’m seeing less enthusiasm for 10bit 4K All-I at 400Mbps with every post, that’s not to say that it’s a bad thing, it’s just that it may be overkill……….Are we building a consensus that IPB at 150Mbps makes sense for the majority of our work?
 
Vardalos said:
Are we building a consensus that IPB at 150Mbps makes sense for the majority of our work?
Click to expand...

I don't think enough people have shot 400 Mbps, or even know when to use 400 Mbps, to reach a consensus.

I've played with it, but I haven't made time to properly test it with motion against the IPB formats.
 
I plan to use All-I for narrative films where I'm not needing to roll for long takes and need every last drop of performance and quality from each frame.

For the rest, IPB works really well and no one except us pixel peepers are going to notice the difference.

Also, there is a deficit of good/fast/cheap UHS-II cards out there currently to support All-I. I'm waiting for SanDisk to release guaranteed V90 cards before jumping in, as they've looked after me since getting the GH4 and I can't find fault with their cards, whereas plenty of other brands seem to have QC issues.

Cheers,

Paul :)
 
visceralpsyche said:
I
For the rest, IPB works really well and no one except us pixel peepers are going to notice the difference.
Paul :)
Click to expand...

I would like to see that pixel peeping difference. Static frame and motion frame. What may be the size difference of i-frames (all-i vs ipb)? i-frame size defines the video quality if there is little motion.

I think it is like high quality i-frames and calculated motion vs low quality i-frames and precise motion.
 
Vesku said:
I would like to see that pixel peeping difference. Static frame and motion frame. What may be the size difference of i-frames (all-i vs ipb)? i-frame size defines the video quality if there is little motion.

I think it is like high quality i-frames and calculated motion vs low quality i-frames and precise motion.
Click to expand...
Not sure what you are saying here.

The size of I-Frames, i.e. the compression level of I-Frames is not dependent on motion, at least not directly, it depends on the available bitrate.
I-Frames in an IPB video could be of a higher quality than the same video in All-Intra mode due to exhaustion of available bits, this is especially true in videos with little motion.
 
Vardalos said:
I’m seeing less enthusiasm for 10bit 4K All-I at 400Mbps with every post, that’s not to say that it’s a bad thing, it’s just that it may be overkill……….Are we building a consensus that IPB at 150Mbps makes sense for the majority of our work?
Click to expand...
No, not at all. I prefer to shoot narrative work at ALL-I for an optimized post workflow. I shoot events with IPB `cause I need long recording times and push a lot less in post.
 
joe12south said:
No, not at all. I prefer to shoot narrative work at ALL-I for an optimized post workflow. I shoot events with IPB `cause I need long recording times and push a lot less in post.
Click to expand...

Now that V90 cards are coming down, I think you've laid out a good plan forward. All-I for the good stuff and IPB for events.
 
Last edited:
Vesku said:
I finally updated my GH5. I tested all-i vs ipb. Here are static and moving frames from 4k 10bit all-i 400Mbs and 4k 10bit 150Mbs ipb converted to max quality JPGs. Static frames are quite identical but the moving frames are cleaner in all-i.
I posted the exmples here:

https://www.dpreview.com/forums/post/60301188
Click to expand...

Thanks for posting. To my eye almost imperceptible difference on the moving samples. First I think I see a difference then I don't.
 
You must log in or register to reply here.
Back
Top