GH2 - 2.5fps AVCHD 24h Timebuster/TimeHBusteR 2.0 settings - The day is not over!

duartix

Timelapses can be amazing, beautiful, strange, provocative. And FUN!

The day I discovered the GH2 could do 2fps I got so excited, I loaded up a full battery, put the camera on the balcony and aimed it at the shopping mall door. I got over 3+1/2h of Cinema 24p footage (@2fps) and after speeding it up 15x (to 300fps) in VirtualDub I almost couldn't believe my luck. This is what I got and I have yet to grade it properly:

Since that event, I wondered about the best way to patch up the camera to capture them. MJPEG was the obvious candidate. It's a Full Intra codec and it can be setup to capture just 2fps. That means absolutely no wasted frames unlike AVCHD, so MJPEG must be great, but there was a caveat: even with stock settings there was a capture time limit that was well below AVCHD. How is that possible since it's only capturing 2fps, the bitrates have got to be frugal? Well, it happens that there is a 2GB filesize limit within MJPEG so even after taking Driftwoods Quantum 100-50's MJPEG settings and cutting the quality to take half the bitrate, you will end with less than one hour of footage. Compare that to AVCHD and you see what got me started.

Then there was the IQ issue. I stumped into this: http://sonnati.wordpress.com/2010/10/19/h-264-for-image-compression/ which made me start another full set of tests (http://www.personal-view.com/talks/discussion/2111/gh2-mjpeg-resolution-findings-and-myth-busting/p1) in which I found that unlike most crazy MJPEG resolutions that only seem to be 720p upscaled, the one used on Driftwoods patch was the best that ATM can be extracted from MJPEG and that even rivals with AVCHD. That test ended up on a tie, so in my book AVCHD wins for flexibility and recording time.

At the same time these tests were taking place, I fought a lot with the stock settings trying to make a patch that used GOP 13 and yes, that's the frequency at which you see major block changes when you set the shutter speed to 1/2s. This GOP13 would insure excellent quality as it makes the result Full Intra once you trash the needed frames. But the frames and limits and buffers and modes were giving me all kinds of speed messages Hell. By that time I also discovered that the null B frames that encoded no changes at 2fps were much larger than P-Frames which also appeared a lot more frequently than I wished they would. Null B-Frames were 2.5x bigger than P and when there was movement, the P-Frames would take as much space (and sometimes even more) than I frames, so I had to do something about it to make it more efficient. Meanwhile Nick @Driftwood and @LPowell were kind enough to share their knowledge and let me in on a few tricks. These allowed me to suppress B-Frames altogether and starve P-Frames to death.

So here I was armed with a bazooka and a missile and I couldn't even fire a pistol. Then a few days ago @Balazer came out with his Cake patch featuring a completely VBR approach through a Constant Quantizer as if he couldn't care less about bitrates, limits or buffers or all the stuff that was putting me down. I tried it a 2fps and I couldn't believe how well it worked. And since the frame rate I'm using is ridiculous when compared to normal video modes, there are no speed/buffer/spanning issues even if you have the slowest SD card in the business. In this case, this VBR aproach is a no brainer and the cherry on top is that you get Constant Quality at Quantizer 22 which is very very good IQ!

So, hands to work, bake the Cake, mix Driftwood's MPEG settings, apply the P/B-Frame tricks and it was working on the third try. :O

It could have been a lot faster as I tried to best the 60i mode and leave all other modes alone. It ended up being the wrong choice because the 60i GOP isn't sticking and probably the Interlacing is making too big P-Frames around the I-Frames. In the end I managed to use tweak 24p for timelapses and the frame usage when there is no movement is a beautiful set of I-Frame spikes with all else almost null. 60i is now GOP13 on I/P frames exclusively and the P sizes look generous, so the starved P-Frames on 24p don't seem to affect it. 720p modes seem also unaffected. In the end this patch tries to give you the best 2fps and still leave you with 2 very workable modes without repatching: 1080i60 and 720p.

Thank you @balazer for baking Cake.

Thank you @drifwood for the MJPEG settings and no B-Frame trick.

Thank you @lpowell for telling me how to starve the P-Frames.

Thank you @cbrandin for StreamParser 2.4 which gave me wonderful new numbers.

.

VERSION HISTORY :

1 - TimeBuster 1 - First Release

Based on Cake v1. It uses GOP13, a Constant Quantizer VBR approach without B-Frames, minimum P-Frame size and quality.

2 - TimeBuster 1.1 - with 2 significant changes:

Now based on @balazer 's Cake 1.1, it has switched to use the Rate Control mechanism from @LPowell 's FlowMotion 100.
720p mode was broken due to use of a very low quantizer (12) which lead to skipped frames in 720p. It now uses QP=20 on both 720 & 1080 modes which will have an observable effect in both bitrate & quality even though it still gives you 24+ hours on 64GB.

3 - TimeBuster 1.2 - with 1 significant change for a fully featured 720p mode:

Since @balazer 's Cake 1.1 is now heavily based on @LPowell 's FlowMotion 100, I decided to copy LPowells 720p settings since they will probably play better with the other modes. The only difference on 720p is a constant QP which had to be lowered yet from 20 to 21 because it still dropped frames on torture charts and/or sometimes present speed issues at very high ISO (>3200) on Class 10 SDs.

4 - TimeBuster 1.3 - with a lot of significant changes:

Ported to PTTools 3.64d
Savings on P-Frames is now accomplished by using a Scaling Table that is full of 0xFFFF for 1080p modes.
Quantizer boundaries were removed since they are now much more dependent on the new Quantizer Tables.
Redesigned for merging, these definitions try to be as independent as possible from their re-packaging definitions (in this case FlowMotion 1.1) and are published in two flavors: Timebuster 1.3 Base (for merging) and Timebuster 1.3 FloMo (merged with FlowMotion 1.1)

NOTE1: Timebuster Base 1.3 was merge tested with @LPowell 's FlowMotion 1.1, @Balazer 's Cake 1.2 and Nick @Driftwood 's Quantum X (v2) with minimal impact on all modes except for 1080p. I recommend FlowMotion or Cake for the simple reason that the 1080p24 Frame Limit is set to a much higher value in these settings. In fact, if you care for nothing other than the 24p timelapse IQ and don't mind impacting some of the other modes, you should set it to 10,000,000.

NOTE2: Merging Timebuster Base 1.3 with other settings is easiest done by loading Timebuster Base 1.3 first and then Alt+Clicking your choice of merger.

5 - Timebuster is now comprised of 2 main variants:

a) 24h Timebuster 2.0 - with the following changes (from TimeBuster 1.3):

The GOP for 24p is now married to 2.5fps. The bitrate hit is a little above 20% but the recording length can be easily compensated by raising the QP. As a benefit it's still very close to 2fps and it's the closest to a 360º shutter we can get on 24p.

b) 24h TimeHBusteR 2.0 (NEW, uses HBR 30)

Similar philosophy to 24h Timebuster 2.0 Base but as the name suggests, uses HBR mode instead of 24p. There is another difference, P-Frame quality is fully preserved as this kind of frames is used for one of the interlaced slices on Key Frames.
GOP married to 1/2.5 on HBR 30p (NTSC) and to 1/5s on HBR 25p (PAL). The big benefit is that both are using a 360º shutter. Uncheck "1080i50 and 1080p24 GOP Size=5" before merging if you don't plan to use Shutter Speed (SS) =1/5s and want to better preserve the 24p mode and HBR 25.

Both variants are now:

Using B-Frames. There is a 6% penalty in bitrate usage but this was necessary to make sure the QP was respected. I believe the gain from setting the QP at will is just too great to ignore, since not everyone wants to record for 24h straight. QPs as low as QP=16 were tested and theoretically it could go as low as QP=12 @ISO160 when merged with FlowMotion 1.11 on regular tripod mounted footage.
GOP married to 2.5fps. The bitrate hit is a little above 20% but the recording length can be easily compensated by raising the QP. The reason behind this was getting as close as possible to 360º.
User definable GOP - GOP can be set at will according to intended SS, use the attached "GOP-SS-table.png" to look up the right value. Defaults settings are SS=2.5fps and GOP 1080i=10 for 24h Timebuster 2.0 and SS=2.5/5fps and GOP 1080i50/i60=12/5 for 24h TimeHBusteR 2.0.
User definable QP - QP can be set at will according to desired IQ/recording length. I've gathered quite an amount of information so that in the future I can make an Excel tool to estimate the best QP from Card Size/Footage Snapshot/Desired ISO/Recording Length. Default settings are Q24/23 for 24h Timebuster/TimeHBusteR.

As before, there is the Base Release to be merged at wish with popular settings (load Base first in PTTools and then "ALT+Click" your choice of merger) and there is also the FlowMotion Release (already merged with FlowMotion 1.11). I've merge tested against @lpowell 's FlowMotion 1.11, @balazer 's Cake 95, and @bkmcwd 's GOP3ZILLA and no undesirable effects were observed. The key is using a VBR settings definition with a very high "1080p24 Frame Limit", especially if you are going for a low QP (<22).

(edit) If you want to merge with Nick @driftwood 's settings, I recommend going for Cluster v1 as shines in being both a long GOP set and also having the biggest "1080p24 Frame Limit" of all the sets I know.

The day is not over. :)

duartix

Here's what it looks like on 24p @2fps on static footage. Almost as good as 2fps MJPEG would look. And the only impairing I observed (on 60i) was the long GOPs, but with good P-Frame size. The beauty of the patch is that you can now easily trade IQ for recording length with very little changes to Cake (as I did). Probably all you need is to balance the frame limits with the QP.

This is how 24H looks on a very detailed and colored chart. Average Total Bitrate = 7.012.276. On real life and detailed footage but with less texture, I got under 5Mbps (on 24H).

On a 64GB SD this translates into 29 hours.

THE DAY IS NOT OVER.

Tommyboy

This is awesome! naf SAID!

eyefi

@duartix, are you in lisbon?

RRRR

Impressive stuff!

duartix

@eyefi:

Yes, busted!

Don't tell me you are the guy with the girl... :)

eyefi

@duartix, ok I don't tell you. keep up the good work. Maybe we'll talk soon.

tida

@duartix does it mean you would have created the perfect patch to combine timelapse at 2fps (which would give factor 13x as a base) with slow motion at 60i transferred to 24p (which would give factor 1/2.5 as a base) and all at 1080p.

If you can combine it like I did it as shown below it would be an awesome timebuster with extreme possibilites of dynamic range...

duartix

@tida : Yes, well to be honest, keeping 60i was an accident. I was going for an all round patch trying to preserve 24p for regular shooting and dedicating just 60i for timelapses. But as I wrote before, something I didn't foresee made me reconsider: 60 is "i", which means that at 2fps when the footage changes are recorded by the camera from t=#0 they will be spread across 2 fields at frames #30 and #31 and to encode these changes properly in Full Intra, you would need GOP1 which defeats the purpose of getting long record times.

In the end, I'm glad this happened because this is a timelapse dedicated patch and this way you can preserve the modes that work on extreme timing ends. 720p60 is preserved, 2fps @1080 has great recording times and IQ, and you still have 1080i60 with very good quality and resolution. I've tried to set 60i GOP length to GOP1 but as I said before, it doesn't stick, and the camera goes with 13 in that mode. (edit) Stupid me, (I was testing PAL, that's why it didn't look like it sticked) (/edit) It's not too bad but it's not ideal either, so if someone has an idea on how to change it, please let me know as I'd like to try it!

It's all about choices, as much as I'd like to do an all round patch, the focus here is on timelapses and if you really need 24p at the same time, well then you might as well use another SD card with a better 24p patch on it. Provided you have 2 cards, you can change the patches on the field as many times as you need, all it takes is 3min at a time and a full battery (two things that timelapsers usually can spare :D ). Timelapses can be done with amazing quality in the smallest & cheapest cards, so I think the card is not an issue.

Hallvalla

@duartix A couple questions:

Does this patch actually record @ 2fps in AVCHD? or does the camera still record 24fps and you have to remove all of the duplicated frames? I couldn't find in the pTool settings where you switched 24fps ---> 2fps.

Do you still have to use a 1/2 second shutter, Manual mode + manual focus?

Could you write a simple workflow from start to finish (camera settings to post) about the timelapse video you've posted?

This is very exciting, thanks for you time and efforts!

~J~

duartix

@Hallvalla : Yes, you still need to put the camera into MF so that you can then slow the shutter speed manually to 2fps. The camera will still record 24fps, the thing is that most of those frames will be subjectively NULL (they still occupy some minimal space) as you see in the first StreamParser capture. When there is significant movement however, some of the P-Frames surrounding the I-Frames will occupy some noticeable space.

I'll try to capture something today and post it here. I'll try to post the workflow too.

Rafa

@duartix : How you get the 2fps in camera? Sorry but I don't find the way.

Vitaliy_Kiselev

@duartix

Real approach to this must be ability to change encoder so it'll skip the frames and you must be able to specify how many you want to skip. It'll also require to completely turn off audio.

duartix

@rafa As it was said just before, you need to be on Manual Movie mode, and set the camera to MF. Only then can you reduce the shutter speed below 1/25(30)

@Vitaliy_Kiselev : That would be really great as P-Frames sometimes still get hungry and audio also can account for up to 5% at the moment, any chances those would be possible?

Until then, there is a saying here in Portugal that goes like this: "He who doesn't have a dog, hunts with a cat."

duartix

@Hallvalla:

1 - Camera top Dial must be set to M Movie, video mode must be set at 24p (cinema) and focus to MF, with the rear dial set shutter speed to 1/2s (pressing the rear dial against the camera changes control from A to S).

2 - Record.

3 - Copy MTS file to disk .

4 - Open VirtualDub (make sure you have the DirectShow filter installed http://forum.videohelp.com/threads/286901-VirtualDub-Now-Has-a-Directshow-Driver-Plugin )

5 - Go to File Open (make sure that in "File of type" you select "DirectShow input driver") and open the file.

6 - On Video menu select Frame Rate. There are two options on panel Frame Rate Conversion:

a) Decimate by 12 (null frames will be deleted but you won't get the speed effect)

b) Convert to 300fps (accelerated)

7 - On Video menu configure the Compression codec.

8 - Save file as...

This is the easiest workflow I can imagine (on my video I applied some basic filters (menu Video | Filters) and used the Blend option to blend a B&W Fade Out and Fade In.

The guaranteed better way to do it (to make sure you are only selecting I-Frames) should be with AviSynth and this post should give you an idea: http://www.personal-view.com/talks/discussion/1559/avchd-1080p-time-lapse/p1

There are also blending options on After Effects but I know very little about it.

Hallvalla

@duartix Thanks. Hmmm, just found out VirtualDub runs on Mac OS X through Wine (like pTools). Will give it a shot.

duartix

TimeBuster 1.1 is out, featuring 2 significant changes:

Since this is based on @balazer 's Cake 1.1 it has switched to using the Rate Control mechanism from @LPowell 's FlowMotion 100.
720p mode was broken due to use of a very low quantizer (12) which lead to skipped frames in 720p. It now uses QP=20 on both 720 & 1080 modes which will have an observable effect in both bitrate & quality even though it still gives you 24+ hours on 64GB.

The day is not over. :)

tida

@duartix thanks, looking forward to use it. If one would like to use higher shutter speed than 1/2s would the patch keep it's performance by just lowering the GOP. Would the calculation in below mentioned macro fit?

http://www.personal-view.com/talks/discussion/1559/avchd-1080p-time-lapse#Item_61

StepSize1 = abs(ShutterSpeed_s*FrameRate_fps)+1 //For your Patch StepSize1 == GOP13 at 1/2s

duartix

Let's see what are the most desirable SSs:

SS=1/2s then you get: 1/2*24+1=GOP13
SS=1/4s then you get: 1/4*24+1= GOP7
SS=1/6s then you get: 1/6*24+1= GOP5
SS=1/8s then you get: 1/8*24+1= GOP4
SS=1/12 then you get: 1/12*24+1 = GOP3

(EDIT) THIS LOGIC ISN'T WORKING. Look further in this thread: http://www.personal-view.com/talks/discussion/comment/39686#Comment_39686

Apparently the formula guarantees that you'll have a new I-Frame that is pretty much in synch with your later frame sampling rate and that the rest are tiny null P-Frames to be discarded. If I were in your shoes, I wouldn't go for other funny SSs like 1/5 or 1/10 since 1/4 and 1/8(12) are very close and will work even without scripting tools.

In Timebuster 1.1, I've lowered the QP from 22 to 20 so we get a little better IQ but still get around 24h. In principle the patch will keep it's IQ performance but it will pay for that in bitrate and recording length. All other things equal a 1/4 SS will spend 13/7x the bitrate when compared to 1/2s, so you'll get just short of 13h of footage on 64GB. At 1/8s it's well over 7h. Do take into account that this will always vary with footage content and mostly with ISO.

Having said that, this is a game of just matching GOP with SS. The beauty of recording footage at lower SS is that it's much less demanding on the Camera CPU, on the codec and on the storage card. If you have recording length to spare you can even be adventurous and lower the value of "Quantizer for 1080 modes", but remember that if you lower that QP by 6, you'll end up with approximately 2x bigger frames and then you might already be hitting the ceiling on the 1080p24 Frame Limit.

Q20 even though it is the default stock setting, it is very good quality already IMHO and if the stock GH2 wasn't so limited in buffers, frame limits and bitrates, it would be spitting out so much quality we wouldn't be using any hacks...

balazer

Lowering the QP by six doubles the precision of the coefficients. It doesn't necessarily mean the coded frame sizes will double.

duartix

OK, thanks for the clarification. I'll perform some tests to compute the impact.

tida

@duartix what could be the reason that I find with your patch that especially under stress the 13st-1 frame answers with high bitrate.

duartix

I believe it's a common thing among all patches I tried. I don't know if it's the camera normal behavior saving up bitrate to build up a buffer for rainy days, but I know there's a parameter called "Initial quantizer". I suggest you lower it from 20 to let's say 18 and see what happens.

tida

@duartix I know there was a big discussion with @balazer about GOP 12 or GOP 13 for 1/2 sec at 24p but is that really 100% sure to use GOP 13. Why has Panasonic used the standard parameter GOP 12. For 25p I understand that it must be GOP 13 but... maybe I should try your patch with GOP 12 and look how it behaves under stress. What is your practical experience...

13/23.976 = 1/1.844 > 1/2s == go, but why not the next....

12/23.976 = 1/1.998 > 1/2s == go

12/25 = 1/2.083 < 1/2s == no go

13/25 = 1/1.923 > 1/2s == go

@Vitaliy_Kiselev is there already something which you know about standard GOP settings for 25p from the new firmware.

In a GOP12 at 24p case I would have to implement for my macro

StepSize1 = round(ShutterSpeed_s*FrameRate_fps) == GOP

which sounds more logic for me...

duartix

@tida : Logic has nothing to do with it... :) Otherwise your simple and logical formula should work!

At first my thought was that the parameter AVCHD GOP Length in PTTool was counting the I-Frame twice unlike the standard definition of GOP: http://en.wikipedia.org/wiki/Group_of_pictures but that's not the issue, the parameter conforms with the definition. The issue here is that the camera @2fps spits out the frame changes once every 13 frames and that's why we're using GOP13 to stay in synch (and I've just loaded a 30min MTS file in StreamParser and confirmed that it still stays in synch in the end). If you use GOP12 you'll see a big P-Frame travelling away from the I-Frames and passing by an I-Frame every 12 GOPs.

Now this leaves me weary about the other shutter speeds and I'll have get to check their GOP structure.

BTW, Panasonic chose GOP12 most probably to maintain a consistent I[BBP]+ GOP structure.

tida

@duartix thanks for your explanation, you have done a very good job. It sounds like a proof for me. One explanation could be that 50i and 24p has to share same GOP size.

duartix

BEWARE @tida and TIMELAPSERS !!!

Analysing the framesizes (and also checking for content) of footage shot with Shutter Speeds (SS) below 1/25 reveals that the "Real" Framerate is NOT 1/SS as should be expected!

SS - Frequency of new frames - Ideal patch GOP

1/2s - 1/13 - GOP13

1/2.5 - 1/10 - GOP10

1/3.2 - 1/8 - GOP8

1/4 - 1/7 - GOP7

1/5 - 1/5 - GOP5

1/6 - 1/4 - GOP4

1/8 - 1/4 - GOP4

1/10 - 1/3 - GOP3

1/13 - 1/2 - GOP2

1/15 - 1/2 - GOP2

1/20 - 1/2 - GOP2

This means that when shooting with SS=1/20s, the content of frame1=frame0, frame3=frame2, etc. There is no new content every 1/SS frames!!!

ten

Hello,

I been lurking here for a short while after getting my gh2 last week. Tried a few patches but quite interested in doing time lapses so am looking at Timebuster. Sorry if these are newbie questions.

Mjepg mode is already at 2fps I believe so all you need to do is speed it up in your nle ? But avchd mode you have to bring ss down to 1/2 for 2fps?

Also this is based on cake 1.1 so was wondering the difference from simply reducing ss to 1/2 when using the cake patch to doing the same in timbuster? Are there differences in the patches for avchd?

Again sorry for newbieness!

Cheers!

duartix

Yes, MJPEG frame rate is fixed and set in the patch to 2fps and yes, in AVCHD you do need to set the SS to 1/2s to marry GOP13.

The major difference will be (and this is an estimate since it will depend on the content, on the quantizers that are different and also a lot on the P-Frame size reduction) that you'll be able to record around 5x more content because I've set the 24p GOP to 13 (instead of 3) and because since I won't be using the P-Frames, I've totally trashed their quality in 24 p to T4 (instead of T1). I've recorded a timelapse with 1.1 similar to the one I posted on the first thread that would record for ~22h on a 64GB SD card. At night! Oh wait, that's impossible at this latitude... ;)
The second difference is that I reduced the Quantizer for 1080 modes to 20 (from 22 in Cake). That should give you a perceivable boost in IQ. All that happens while still managing to keep the average bitrate below 5.925.925bps which is the "Magic" number that gives you 24h on a 64GB SD, the major objective of this patch.

IIRC other than these two differences the patches are identical when it cames to AVCHD.

BTW, the Quantizer for 1080 modes is the single most important parameter in this patch. At 2fps it could easily give you either half or double the time if you are willing to trade recording time for quality.

The other is 1080i50 and 1080p24 GOP Size, and if you don't give a rat about continuous motion flow then by all means, go bananas, put 26 or 39 in there and near double or triple the recording time. Hell, go bonkers and set it to 117!!! You can record for close to ONE WEEK if you also raise a little bit on the quantizer. You'll compromise playback however, and 60i.

(edit) You can't go over GOP 50.

ten

Awesome, thanks for the info. Going to put this to good use on some TL over the next week ! cheers!

tida

@duratix there is the basic expression in rheology: "everything flows"

Might be one time you will see the glass of your windows flowing...

PapaRomeo

Hei duartix, I get this "Unable to set filename on media detector: hr = 80040200" with this Virtualdub software.

Is there a quick fix? Windows 7 home premium 32bit, Bootcamp

Can't the footage just be accelerated in FCPX?

duartix

@PapaRomeo, I don't know FCPX but I'd bet it can do something as simple as accelerating the video as can After Effects. I just gave @Hallvalla this workflow because it was the simplest I could imagine and it used only one piece of software. I can't tell you how VDub works under Wine because I use it natively on Windows 7 64bit. Anyway a google search for that error message leads here: http://forum.videohelp.com/threads/286901-VirtualDub-Now-Has-a-Directshow-Driver-Plugin

tida

@duartix as far as I see it from your GOP results it would make sense to create an additional GOP 8 patch to cover most of the other possible frequencies of new frames. Then it should be allowed to have P frames in 2nd, 4th position. Or only one universal 8x13 = GOP 104 patch for all with good quality...

duartix

@tida: That would make sense had I not totally trashed P-Frames IQ with "1080p24 T2=1080p24 T4". But you do have some point however. I could give it a go at using B-Frames and trashing them while keeping P-Frames with more balanced IQ.

If the bitrate wasn't compromised at QP=20 and there was more demand for SSs other than 1/2s, I could make an alternative patch with GOP104, but analysing the traditional GOP structure I BB P BB P (...) it looks like there will be a lot of P-Frame waste on a lot of SSs. This "traditional" I(BBP)* structure is best fitted for GOP3/6/9/12 which only suits 1/10s. If I could get the GOP structure to behave like I(BBBP)* (which ATM I don't even know if it's possible) it would suit a few more cases : 1/6s, 1/8s and 1/3.2s. I'm starting to think that it's not worth it, because it will waste a lot of bitrate on all others... :(

Probably the best option for specific SSs is to have a set tailored patches for specific SSs on different SD cards and upgrade the FW on the field according to the footage demand. Just to give you an idea, on a GOP8 patch for 1/3.2s, I estimate the bitrate to be around 50% more at the same quality when compared to the GOP13 for 1/2s. I also don't believe there are nasty consequences to taking my patch and just suiting "1080i50 and 1080p24 GOP Size" to your current needs.

tida

@duartix the enormous breakthrough is of course the 2fps way.

Your patch is good to speed up to 13, 26, 52, 130, 260, 520, 1300 times. 1300 times means nearly 21 hours results in one minute whereas 100frames have to be blended.

With 24fps as usual you can speed up easily 2, 4, 10, 20, 40, 100 times without wasting to much memory. Only 100frames would have to be blended to result in one frame at 1/25s. 100 minutes would result in 1 minute.

Even the quality (especially noise reduction takes place) with standard settings is incredible if overlay method or temporal soften method of AviSynth is applied for blending.

I only want to say that above values display how you feel time changes within a day. That is not in a linear scale, it's in logarithmic scale.

(Divide a logarithmic sale in 3 and you will end up roughly with numbers like 1, 2, 5, 10, 20, 50, 100, 200,..... Think about your money - cover what you have to pay with the lowest amount of pices in your pocket)

You cover already "short", "medium", "long" in an economical way...

duartix

Thanks.

I should also say that the quality is already amazing @2fps... :) even without frame blending. IIRC I had to reduce aperture to about F5.6 on my 45mm F1.8 Olympus to keep the highlights on the timelapse I posted... and that was at ISO160 when the ISO test thread shows that ISO640 is the least noisy!

PapaRomeo

Thanks Duartix for the clarification. The patch is very useful.

duartix

@PapaRomeo: You're welcome.

Next on the shortlist is:

balancing 720p
trying starved B-Frames
trying very long GOPs

AmandaNL

Thanks for the effort @duartix I am planning to test the patch tomorrow. It will be my first time-laps so hope it goes all well :-)

duartix

720p is fixed now (it's now based on FlowMotion with a 22 QP as it still dropped some frames on torture charts), running the other tests and I hope to have it published in a few days.

dannywong3

SO~~there is no chance to combine any driftwood's pacthes with your patches together??? I'm afraid there are some config that use for timelapse and high bitrate patch....?

duartix

Totally different targets, objectives, priorities and means.

@dannywong: Sorry, almost zero possibilty. Mine is mostly about saving bitrate, @driftwood 's is about spending. Mine is mostly about long GOP, driftwood's about short GOP. Mine's about VBR, his is is about CBR (if I'm not mistaken). On top of it all, we both had to concentrate on 24p, so they are as incompatible as you can imagine.

Or were you talking about the 720p modes?

(edit) BTW, don't get me wrong, this is a mainly VBR patch and will use all the bitrate available to it while keeping the quality of the I_Frames at Q20. And the bitrate caps are not even close to stock settings, it's 80Mbps and 70Mbps IIRC so at 2fps it's as if it wasn't even capped with the QP as the main controller.

AmandaNL

Did a small test and it went well. Recorded 30 minutes (836mb on card) and speeded it up to 3.000 percent in Premiere Pro and it looked very nice. Will do tomorrow another test and upload some footage.

duartix

Releasing Timebuster v1.2

TimeBuster 1.2 is out, featuring one significant change which now gives it a fully featured 720p mode:

Since @balazer 's Cake 1.1 is now heavily based on @LPowell 's FlowMotion 100, I decided to copy LPowells 720p settings since they will probably play better with the other modes. The only difference on 720p is a constant QP which had to be lowered yet from 20 to 21 because it still dropped frames on torture charts and/or sometimes present speed issues at very high ISO (>3200) on Class 10 SDs.

Tried a few dead ends:

Starving B-Frames - didn't work even after messing with scaling tables. B-Frames at 2fps were still twice as big as the P-Frames it is still using.
Very LOOOOOONG GOPs - The biggest GOP that I could get to marry 2fps without freezing the camera was 50. If you are willing to sacrifice three out of four frames every 2s, use GOP50, I managed to get bitrates a little below 1,900,000bps at ISO160 which could give you 3 DAYS recording time on a 64GB SD. Getting even more would start to require some kind of success in getting rid of the parasite bitrates (like audio) or giving up IQ (by raising the quantizer).

As for the current settings it still will typically give you 24+ hours at 2fps with excellent IQ on 64GB.

The day is not over. :)

duartix

Thanks @AmandaNL !

I'm looking forward for see it. :)

duartix

ATTENTION!

"Quantiser for 720 modes" will most probably be set to 22 again in the next release. It turns out I've been testing 720p60 the wrong way all along and it was still dropping P-Frames... :(

tida

At http://www.personal-view.com/talks/discussion/2232/different-times-at-one-location-extended-dynamic-range-solution-for-1080p-videos#Item_1 you will find an easy to use solution to create AVCHD timelapses. All presets are designed for the timebuster patch. It also includes the GOP results at different shuter speeds which @duartix did analyse.

duartix

Cheers, Timelapsers and Timebusters!

Yesterday I discovered I had made a such a stupid slip up when testing Timebuster that it could even compromise my whole approach to the patch set.

To put it simple and blunt, I mistook 30p/60i for PAL and 25p/50i for NTSC. So, while in my PAL tests I thought I was testing Quantizer for 720p=21 for 60p, in reality it was only 50p, and P-Frames were still being dropped in real 60p. It doesn't happen anymore now with Quantiser for 720p=22 at least up until the very last extreme ISOs.

Sorry for that! My fault! But the main objective of the patch wasn't compromised in any way and the solution is simple! :)

(edit) Since it's a very small change and the patch looks very stable already, I won't be updating it until the patched fw 1.1 comes out which will possibly trigger a new version of FlowMotion.

Now, what might be compromised is the approach to the patch set. My first take was tackling GOP just in 60i in order to keep the rest of the modes uncrippled (especially 24p) and having an all round patch that could be left forever in the camera. Since my testing was wrong I and I was lead to believe that the "GOP for 60i"parameter wasn't working at all, I quickly changed the approach to use 24p instead. There were a few benefits in this approach as 60i being interlaced doesn't suit itself all that well to GOP games (because the real content changes at 2fps are spread into 2 frames/fields, and by starving P-Frames I'll be strongly compromising the IQ of one of the fields!). The other big benefit was still keeping the 60i uncrippled, which might be dear to extreme frame rate aficionados.

Does anyone see other benefits/problems in trying that opposite approach by tackling 60i instead?

AmandaNL

@duartix Uploaded some samples. It has 2 sunsets and one highway time-laps, starting in the clip @ 2,05

duartix

Those are great @AmandaNL , the highway timelapse works really well with the soundtrack. But it left me with a doubt, is the motion flow really continuous at 2fps? The tail red light strikes seem interrupted but it might have come from the Premiere speed up, so how is the original footage?

Howdy, Stranger!

Categories

Participants

Tags in Topic

Top Posters