Video Recording Formats AVCHD vs AVCHD Progressive

Since a few years I have had little rigs capable of recording double frame rate videos. More specifically I have a choice of:

1. 24/25 p at 24 Mbps
2. 50/60p at 28 Mbps

I have since been choosing option 2 with a view of being able to slow motion the footage when required,

Out of the recording I create two files a 25 fps and a 50 fps version.

The 50fps version plays beautifully however even on a powerful quad 3.6 GHz iMac I see dropped frames. The mainstream version is therefore the 25p.

The camera follows the 180 shutter rule at 50p so it shoots at 1/100th this gives videos a bit of a staccato feeling in certain scenes so am now thinking of dropping this to 1/50th although this gives me some challenges at time as the camera smallest aperture is f/11 and lowest ISO 160 (I shoot with a Sony RX100 II).

Slow motions is something I use at times for handheld macro to avoid vomiting when I watch the footage later.

It is a pain to have multiple file formats in the same project so I have so far been shooting in 50p all the times slow motion or not.

For my next trip there won't be any macro so am looking at experimenting I am wondering if I should shoot directly in 25p and forget 50p entirely or if it is actually better to shoot at the highest bitrate. I am aware that the H264 codec uses temporal compression so my assumption was that with the same encoding settings the higher the bitrate the better. I have read somewhere else that instead I should shoot 25p directly as there is more information in each frame however I believe this is not correct as H264 does not store full frames like PRORES or other codecs in fact the decrease in file size when compressing from 50p down to 25p is not great maybe 1-20%.

Am I doing the right thing to shoot at 50p 28 Mbps with a view of producing a 25p file or should I go directly to 25p 24Mbps?

I have done several tests and if I just take a snapshot of a frame I can't see any differences between the two that makes me think that unless I slow down the motion and use the extra frame the two iamges are essentially the same in quality, however if I have a fast moving scene the double frame rate should reduce motion artefacts??

Pretty sure your 28 Mbps at 50p is when playing back at 50p, so frame by frame it will have a lesser bitrate than 25p

I have done a bit of digging into the files produced

The 25p have a GOP structure of M=1 N=13 so a full frame is repeated every 13 and the rest are predictions

The 50p have a GOP structure of M=1 N=24 so a full frame is repeated every 24 and the rest are predictions

Both streams have the same number of reference frames in the GOP

So in one second of 25p there are 25/13=1.92 full frames and in one second of 50p 2.08 in both cases with 2 reference frames

In general the 50p format contains more full frames. If I compare the two video bitrates of 22 and 26 Mbps with the additional frames it follows that there is a difference of 8.3% between the number of frames but 18.2% in terms of data rate.

If both files are produced with the same settings the 50p file has actually more information than the 25p based on the number of full frames and reference frames

The difference is however minimal so I would conclude that if you were shooting a static subject there are no substantial differences between the two

if instead there is a lot of changes in a small time interval the format with more frames per second will result in less errors

So in theory 50p recording is the way to go. However once you then decompress and re-encode from 50p down to 25p as output you may as well loose all those benefits

Short version: there is not much in it will keep shooting 50p

That's an awful lot of maths and figures!

Keep it simple, both frame rates have approximately two reference frames per second.... the difference is negligible, not something you'll notice by eye

The compression within the 50p data rate is greater, you're getting 28 Mbps out of 50 frames, as opposed to 24 Mbps over 25 frames.

If you're using 50p to smooth out movement (by slowing it down to 25p), then your resultant data rate is going to be 14Mbps.

If you're render out your 50p footage to a 25p video, you'll be binning half the frames, again data rate is going to reduce to 14 Mbps, so almost half compared to 25p native.

So shoot 25p as the norm, with your shutter at 1/50th, light allowing, move up to 50p and 1/100th when you think the shot warrants it (i.e when the benefit of pushing to slow mo will outweigh the higher compression at 50p)

In terms of aliasing, it really depends on what the camera is doing with the pixels and data. With the FS700, aliasing increases with higher frame rates.... put simply, the camera processor can only do so much, when it's dealing with more frames something else has to give.

Am not convinced as the number of reference frames per second is the same. Inside the group of pictures there is a full frame followed by predicted frame. Each predicted frame is created from previous full frame. There are slightly more full frames in the 50p than in the 25p (though not so many more) so that would make a small improvement in motion prediction (but looking into the numbers actually not that much and this makes me wonder a bit about the whole AVCHD progressive being a marketing scam?)

Anyway this is a not so nice video clip of a fairly static subject I can't see any difference in image quality if I look at a single frame. This confirms that due to the logic of H264 you can't assume double frame rate needs double bitrate and going backwards cuts information. In fact inside the stream the bitrates are pretty much the same once 50p becomes 25p

http://youtu.be/1RYEOU_s1mU

I am going to run a test with something in motion I am starting to think that the AVCHD progressive may be a farce as at 28 Mbps you can't really have more reference frames to significantly improve motion prediction.

Are the bitrates for AVCHD like h264, in that they are variable? So shooting at 50FPS even at 28Mbps with a static scene you wouldn't push the codec anyway and might end up with 15-20Mbps if the encoder decided? So if that is the case, it would not be better to use the shorter GOP of the 24FPS, so recording higher quality 'full frames'?

I would think, keeping the frame rate low, allows better use of the available bitrate encoding limitations, allows more light in, thus keeps the ISO/noise down.

You are correct in fact I find that no matter what you do for fairly static scenes the bitrate does not max out regardless

In general shorter GOP better picture at same frame rate but here it changes however I think the camera decides on ISO before encoding. So it is really the encoding issues that may affect picture quality not exposure.

I mean you get digital noise from encoding not from exposure

I have done another test with a moving ball again I can't see lots of difference other than when playing 50p in slow motions is more fluid however when I then down convert

I have done a little model

In one second of footage at 25p there are 3 full frames and 22 Predicted frames, in one second of 50p 3 full frames and 47 predicted frames

I would not think that the additional 25P frames can be squeezed in just additional 4 mbps if you were at capacity

So that would generally say that if you shoot at 50p you have many more frames closely spaced prediction improves and you can compress more when you shoot less frames you can compress less. All in all seems the result does not change much in general if you have a change of slow motion you should probably shoot 50p. I have done plenty of macro in the past so that suited me fine as sometimes I had to slow down at long zoom.

Also I considered that mixing 50p and 25p clips into a project could be somewhat complicated on the other hand if I had a piece of 50p that I had to slow down the NLE would still convert it into an intermediate format with full frames so that would become 25p in effect.

The problem arises if for some reason I end up shooting 50p for footage I do not want to slow down and end up with a mixed bag

The amount of light in is defined by the shutter speed and the iris, the shutter speeds you able to use are limited by the frame rate. But if, for instance, you set the camera to 1/100th F4 then the amount of light hitting the sensor would be the same, regardless of whether you were shooting 25p or 50p

Am not convinced as the number of reference frames per second is the same. Inside the group of pictures there is a full frame followed by predicted frame. Each predicted frame is created from previous full frame. There are slightly more full frames in the 50p than in the 25p (though not so many more) so that would make a small improvement in motion prediction (but looking into the numbers actually not that much and this makes me wonder a bit about the whole AVCHD progressive being a marketing scam?)

 

Anyway this is a not so nice video clip of a fairly static subject I can't see any difference in image quality if I look at a single frame. This confirms that due to the logic of H264 you can't assume double frame rate needs double bitrate and going backwards cuts information. In fact inside the stream the bitrates are pretty much the same once 50p becomes 25p

 

http://youtu.be/1RYEOU_s1mU

 

I am going to run a test with something in motion I am starting to think that the AVCHD progressive may be a farce as at 28 Mbps you can't really have more reference frames to significantly improve motion prediction.

Have a look at something that has shadows and highlights, compare the pixallation in the blacks, fringing in the lights...

Does Sony AVCHD codec use B Frame?

I read that size of P frame is approximately 50% of an I frame.

So :

25p : Maximum bitrate/ I frame =24Mb/14 (3 full frames and 22 Predicted frames)

50p : Maximum bitrate/ I frame = 28Mb/26.5 (3 full frames and 47 predicted frames)

Is that correct? Any comment?

Stuart, I agree completely with you.

My reasoning behind my comment, is that to keep the shutter angle at 180, for a 50/60fps vs 25/30fps you would have to double the shutter speed (so losing a stop) - so to compensate you would either have to open the aperture or increase the ISO to keep the exposure correct. Hence my comment about running a lower fps and so keeping the ISO low(er).

Thats what I believe anyway!?

Yes there is fringing on the tree on the background on the right side. Same crap for both formats. Shadows and highlights pretty much the same too

I have an updated clip with some motion in it. Now I see the 50p maxing out but still I can't see any differences whatsoever

https://www.youtube.com/watch?v=XDUG7E-TWxY&feature=youtu.be

Does Sony AVCHD codec use B Frame?

I read that size of P frame is approximately 50% of an I frame.

So :

25p : Maximum bitrate/ I frame =24Mb/14 (3 full frames and 22 Predicted frames)

50p : Maximum bitrate/ I frame = 28Mb/26.5 (3 full frames and 47 predicted frames)

 

Is that correct? Any comment?

Sony AVCHD only uses I and P frames no B frames

I have read that for H264 P frame is 25% I frame (50% was in MPEG2) so that would make

25p 22Mb/8.5

50p 26Mb/14.75

I corrected as 2 Mbps are used for audio AC3

Still this would make 25p better but I can't see any difference whatsoever in the examples that I shoot

They are all shot at 1/50th as I am keeping this for the final format of 25p

Stuart, I agree completely with you.

 

My reasoning behind my comment, is that to keep the shutter angle at 180, for a 50/60fps vs 25/30fps you would have to double the shutter speed (so losing a stop) - so to compensate you would either have to open the aperture or increase the ISO to keep the exposure correct. Hence my comment about running a lower fps and so keeping the ISO low(er).

 

Thats what I believe anyway!?

Yep, spot on

 

Still this would make 25p better but I can't see any difference whatsoever in the examples that I shoot

 

50p has higher compression, 25p will therefore be better. Whether the naked eye can tell the difference is an entirely different thing

Don't know if there are people old enough to remember VHS anyway over 10 years ago I converted a bunch of tapes into DVD I was using motion JPEG which was really 25 JPEG pictures per second. The files were huge and there was a direct relationship between bitrate and quality. In the h264 world things are different as you have frame prediction. In general if your compression settings are the same more data equal more quality so a higher bitrate is always better.

However this may or not be relevant to what you use it for. In general I find underwater video not really that action packed things more quite slowly unless you have a swarm of fish schooling and hunting. If I was able to play 50p properly probably that would be best if I had a sport event or similar while for landscape type of use or slow moving objects 25p is good enough. What am skeptical is that the increase of bitrate seems limited. I work on the basis that you can see a difference in quality by eye than it's worth doing. Looks like the AVCHD progressive doesn't have enough horse power to really look much better. Am going to convert the files into an intermediate codec with no temporal compression to see if I can see any difference

I just converted the two clips with the ball trimmed to the same duration into an intermediate codec and guess what they size of the 50p is about double of the 25p.

No difference whatsoever in the quality of the various frames that I can see at naked eye.

If I convert the same 50p files down to 25p I get almost the same file size I would expect from a native 25p

When I look at bitrate the 25p file is now 69.3 Mbps with Bits/(Pixel*Frame)=1.337 and the down converted 50p file now in 25p is 71.7 Mbps Bits/(Pixel*Frame)=1.384

3.5% improvement on bitrate and 3.5% improvement on pixel density per frame of the 50p which confirms there was more information in the 50p at the outset and it is not just dropping frames (it is not possible in H264 to actually do that)

Which really is nothing to shout your momma about as you can't really see differences that which is what my empirical observation was about

Conclusion (for what I am concerned and based on my own understanding of the subject):

1. If you record in 25p you have a little degradation of quality that you can't notice however now all programs can work with your files and edit them no problem and much faster

2. If you record 50p you spend a bit more on the memory card and have a heavier payload in terms of processing time, but you can do slow motion much better if you need to

I am going to try and post this in the sony community forums but am not very hopeful I will get any more insight

Your subjects don't have a great deal of moving detail, so the bit rate isn't going to be pushed. Try throwing a fairly complex image in front of the camera, e.g back lit water droplets hitting a matt black surface, where you're capturing loads of movement, and the cameras trying to cope with a full range of light and exposure, you may see 50p struggle a bit when you look at each individual frame.

If you take a 50p clip and render it out as 25p, your video file will have half the amount of frames. This will be done either through interpolation of adjacent frames (which can result in a fairly messy output) or by dropping frames. It depends on how you've set your NLE up.

If you render a clip out, the bit rate will be defined by the rendering process, the codec used and the parameters chosen. Throw your AVCHD clip through premier and then output as DNxHD, you'll end up with a 180 Mbps data stream. However you're not going to increase quality.... you can't add data in through the rendering process. As for your intermediary codec, no surprise there at all, each looking at each frame, the 50p footage will have twice as many frames as the 25p.

AVCHD is and 8 bit format and highly compressed. It does a great job, which is why it has become so popular, however it does have it's limitations. 8 bit means the image will struggle in low light and highlight, as priority is given to the mid tones where the majority of the important information is assumed to be. Re the compression, as with JPEG vs raw, if you get it right straight out of the box, and don't need to make adjustments, then you'll end up with a pleasing image, however as soon as you start to try to push the image with any form of grade/colour adjustment, you'll very quickly find yourself hitting problems. But in the general case for an untouched clip, it's unlikley that you're going to notice the difference between 25 and 50p.

In terms of bit rate, put it into perspective. BBC HD is broadcast at around 10 Mbps, and is viewed on a screen much larger than your PC monitor. You'd be hard pressed to notice, picture quality is still good... but you can be assured that the initial delivery was way higher than that in order to give them the latitude to pull the image around and get the look they wanted (minimum delivery for us for broadcast is DNxHD at 180 Mbps or ProRes at 185).

If you are shooting 50p, and then retaining the same frame rate in your end delivery, then there will be a minimal increase in quality when compared to 25p. However if you bring the 50p down to 25p, then you will be faced with either dropping frames and losing half'ish of the data, or interpolating frames and quite possibly getting a smeary image as a result.

I would suggest that you approach each shot individually, and choose the appropriate settings. If it warrants slowing down in post, shoot 50P. If it doesn't, shoot 25P.

Pretty much my conclusion Stuart. I think the whole AVCHD progressive spec is a bit of a hoax.

From what I understand time ago camcorders used AVCHD at 17M then other started coming out with 24M mp4 and AVCHD 1.0 was born basically to burn to bluray so it has

50/60i and 1080/24p 24M.

Panasonic who for some reason was lagging behind the 24p 24M came out with 50p at 28M and this was then made into a spec later on but it all seems a bit of a marketing plot to me

What really makes me laugh is the instruction manual

60/50i 24M shoots at the high picture quality of 1080p size (interlaced)

24/25p 24M shoots at a high picture quality of 1080p with a movie-like atmosphere

50/60p 28M shoots at the highest picture quality of 1080p

60/50i 27M shoots at the standard picture quality of 1080p (interlaced)

I would object that as the camera keeps the shutter at 1/100 there never is a movie like atmosphere as a starter and it looks the picture quality comment relates only to bitrate

In general I would think that if there are less frames to be compress each single one could have more dynamic range individually (that I believe is what you are saying?)

So I am going to shoot 25p with 1/50th shutter speed this time. Will probably encounter some issues with excessively bright scenes. I have ND filters 2-8 for land user and a red filter (it still takes away one stop) for bright scenes. When it gets really bad I will have to increase shutter speed

I do not see a lot of macro opportunities in galapagos and if I see some fast action I will attempt a few 50p recordings maybe but probably won't bother at all