This post is designed to answer questions about the H264 codec control in one version of Tragic Lantern in a simple and practical way.
After a few posts I made about using bitrate control on the 600D I had a bunch of messages about settings, and development threads were getting busy, so here's a new thread.
Though one could explain these concepts in greater detail, it's of no real practical use to the average user, so here they in plain English.
They are simplified, so lease don't fill the thread with highly technical responses on video encoding, that information is available elsewhere for those who are curious.
WHAT IS IT?
A balancing act of sorts. High bitrate means nicer footage but bigger files. You can adjust the colours more and remove noise more easily. Advanced control of the H264 encoder that tries to make it use as high a bitrate as possible, without fluctuating wildly or stopping recording.
WHY NOT USE RAW?
Raw can be lovely but there are also a lot of compromises (crop sensor, lower resolution, longer workflow) so perfecting H264 is a great way to get higher quality footage from your beloved camera.
H264 is limited because the encoder uses a rather soft image, but these settings will let you get the best from that image.
WHAT DO I NEED?
A 600D, a fast SD card of 45 or 95MBs will guarantee speed, and (currently) this build of 1%'s Tragic Lantern:
https://bitbucket.org/OtherOnePercent/tragic-lantern/downloads/autoexec.bin-bitrateback
The build is a little older than ones you may be used to so a few things are missing or different. GOP and slice control aren't in the new builds yet but may be at some point, and when they are let me know and alter this thread.
Some other useful things for the 600D are:
The Mosaic Engineering TXi filter, which removes most aliasing and moire, allowing you to make better use of the bitrate. Sometimes it gives Err30 butmost of the time it's amazing. You will still get some moire with exceedingly fine cloth and so on.
The Technicolor Cinestyle Picture Profile and it's LUT, which allows a flatter picture for grading. It's free.
The VisionColor picture profile, which is quite flat but has a nice natural film like tone. It's $6.99. That's a bargain, don't pirate it!
Neat Video denoising plug-in, which can make very high ISOs like 6400 usable when combined with the higher bitrates.
;D 8) THE SETTINGS 8) ;D
SOUND OFF *For maximum image quality at GOP1, don't bother recording sound in camera, deactivate it in Canon menus. Use separate sound.*
IF YOU NEED SOUND use 25p mode and set GOP to 3 instead of 1.
IN SLICE CBR
Lock Slice: Disabled
Min BR: 130
Max BR: 160
Drop by 1: 140
Drop by 3: 145
Taper Rates: Enabled.
IN BITRATE MENU
Mode: CBR
DblockA: -1 (low noise setting)
Dblockb: -1 (low noise setting)
PicPC: 0
GOP: 1 (3 is great too, and saves space)
Bitrate Info: On (I find it interesting)
BuffWarnLevel: 70% (just in case)
BUT WHAT DOES IT MEAN?!!
If you don't need to know, it's probably best to to worry. Hey, the important thing is it works! But here's some simplified explanations:
SLICE
Quality. The numbers used are weird, but not that confusing. The lowest numbers (86, 87) try and use as many bits as possible for encoding. The trouble is, if you lock the slice to 87 and give the camera a hugely complex picture the bitrate goes too high, the buffer fills up and recording stops, or quality has to drop for a while. The second of these situation makes the bitrate "bounce" up and down between loads and 40 or so, which means your image goes pixelly every so often.
Slice control lets the camera raise the slice (drop the bitrate) gradually and in a controlled manner if things get too much.
The settings I've provided here have been stress tested very heavily. They keep things at 87 a lot of the time, but reign it in if the scene becomes too complex. The settings are designed to keep the bitrate below 160 but as high as the camera can keep it without overflowing the buffer in complex scenes.
You can push it harder, but it won't be as stable. These setting have, for me so far, been the most consistant. I recently shot for 15 hours on a fashion video and never had any buffer drops. The average bitrate of the footage was around 100mbps I-frame, which is a lot better than stock.
The buffer can still get too full (resulting in a drop to a very low bitrate and some pixellation) if you shoot a very complex high ISO scene with the slice starting at 87 or so. If you are shooting a complex scene start the slice at 115 or so using the first setting in Slice Control, and the camera will quickly level itself.
GOP
GOP is "group of pictures". To compress more video into less space, some codecs look at one frame, then compare a lot of subsequent frames to it, only storing the bits that have changed. The GOP number is how many are in this group.
So what's the practical result? A long GOP is more efficient, but to my eye tend to look more video like, motion can be a little smeared. I prefer GOP 1, also known as I-frame, as each picture is a whole picture, like a piece of film. The downside here is that file sizes are big, but hey, it looks the best and memory cheap.
GOP 3 is a good compromise and allows sound recording in 25p mode with these settings.
BUFFER
A small memory where the camera stores things before they are written to the SD card. If this fills up, recording will stop. You can use GOP 1 to let frames be written straight to the card, but for some reason GOP 3 seems less hard on buffer. This is reaching the limit of my knowledge. GOP 1 probably stresses the card a lot, but your card will probably be obsolete before it matters. Just format it ater shoots.
Thank you very much for the info, much appreciated.
I have to get a faster card, because when I make the recording it just lasts a few seconds, and I think it's a problem with the sd cards I have. I will get one of ones you recommend.
Best
Xermán
Thanks for this guide! I noticed that topic the other day and tried to look over the last few pages to figure out what was going on, but with all those terms being thrown around, I was way lost. Might give this a try soon and see the difference.
Does this build have the audio monitoring for T3i that was added after the last stable version too? Or is this mode something where you can't do audio recording at the same time because it's too much information and stops recording?
Great theme. I have long wanted casual, what is a "Flush rate"? ( in TL2 ) And GOP = 1 for default bit rate will change in the image?
Flush rate is how fast to write. Gop 1 is ALL-I.. gop 3 is pretty decent middle ground if you still want P frames.
Hehe.. is sraw helping H264 or is canon not using the memory? Sraw should be in both versions.
Quote from: Rick M on July 05, 2013, 03:49:51 PM
Thanks for this guide! I noticed that topic the other day and tried to look over the last few pages to figure out what was going on, but with all those terms being thrown around, I was way lost. Might give this a try soon and see the difference.
Does this build have the audio monitoring for T3i that was added after the last stable version too? Or is this mode something where you can't do audio recording at the same time because it's too much information and stops recording?
Excellent, I'm glad it helped!
I saw many new users in development threads asking similar questions, so I thought it may help.
No audio with these settings I'm afraid. Thanks for pointing it out, I've added it to the post.
I will attempt optimum audio settings too...
EDIT: yes audio does technically work at 25p, but with these settings it overflows the buffer instantly, so I leave it off for best picture...
*double post*
QuoteNo audio with these settings I'm afraid. Thanks for pointing it out, I've added it to the post.
TL 1 you can do audio at 25P I believe... or do wav.. but the wav seems to be synced to 25P in all modes.
Quote from: 1% on July 05, 2013, 06:33:54 PM
Hehe.. is sraw helping H264 or is canon not using the memory? Sraw should be in both versions.
There's no way to check the shoot_malloc size with the TL1 build, or am I missing something?
I tried it with the settings postet in the first post and the picture quality set to "RAW" and "SRAW" (Rebooted, to increase buffer. But I don't know if it worked). I couldn't see a difference. At 30 and 35 frames per second (ISO 3200 to push the bitrate) it stoped with both settings after a few seconds.
But I have a few questions: Why aren't these controls in the new builds? Will they return? Is it possible to set the flush rate with the TL1 build? If not, what's the default?
Flush rate is tied to gop in the first one. Yea, they should return. I'm still dealing w/ raw, etc.
1%, what are your settings with regards to the OP ?
Yes you can do audio, in that it does function at 25p but If you turn it on with these settings the buffer fills straight away, so I leave it off ;)
The image looks great, way better to my eye than the stock one, but the only problem I'm having is that the footage can't be played back in camera, is that normal ?
Yeah that's normal, something to do with chroma subsampling I think.
There is something similar in the canon 550D, Thanks.
Quotethe encoder uses a rather soft image
could you explain that in more detail? I saw somewhere an option to disable sharpening... this looks controversial... do you need to sharpen it or to soften???
Quote from: far.in.out on July 08, 2013, 12:30:42 PM
could you explain that in more detail? I saw somewhere an option to disable sharpening... this looks controversial... do you need to sharpen it or to soften???
I think that he means that the image received by the h264 encoder is soft (it's not a genuine 1920x1080, but reescaled from a smaller size).
For what I know one of the best approaches to sharpening is recording with sharpening 0 and increasing it in post-pro but maybe someone has found a better workflow.
Anyone had time to make a comparison between stock vs custom?
Quote from: Luzestudio on July 09, 2013, 10:50:05 AM
I think that he means that the image received by the h264 encoder is soft (it's not a genuine 1920x1080, but reescaled from a smaller size).
For what I know one of the best approaches to sharpening is recording with sharpening 0 and increasing it in post-pro but maybe someone has found a better workflow.
I have done extensive testing on the sharpening issue. What I've found out, is that the cinestyle picture profile used with 0 in camera sharpening, gives a way to soft image, and if you try to sharpen it in post, you have to use very high values witch produce too much noise on the footage. So, I have learned that if you use +1 sharpening in camera, and then a low value software sharpening, you get a nice clean image, without any noise, and good enough sharpened.
Never, ever, go with the in camera sharpening beyond +1, and in the most extreme cases +2. I recommend either 0 or +1.
Quote from: Critical Point on July 10, 2013, 08:31:52 PM
I have done extensive testing on the sharpening issue. What I've found out, is that the cinestyle picture profile used with 0 in camera sharpening, gives a way to soft image, and if you try to sharpen it in post, you have to use very high values witch produce too much noise on the footage. So, I have learned that if you use +1 sharpening in camera, and then a low value software sharpening, you get a nice clean image, without any noise, and good enough sharpened.
Never, ever, go with the in camera sharpening beyond +1, and in the most extreme cases +2. I recommend either 0 or +1.
Good advice, I will give it a try.
Yes the encoder receives a lower resolution image feed that's been rescaled a few times so it's never true 1080p
I recently started trying sharpness on 1 as well. I've not A/B tested but the idea sounded good to me, since we aren't dealing with raw we could do with a little bit of sharpness "baked in"
I used these settings on another film last weekend. It's looking good! I did find that the camera stops recording in longer takes. Must be the 4GB file limit coming up early...
When I can find source of this build ? I would like to port it to 60D..
Quote from: marekk on July 11, 2013, 11:09:16 AM
When I can find source of this build ? I would like to port it to 60D..
Read the first post: https://bitbucket.org/OtherOnePercent/tragic-lantern/
Yes I know ;))))
but actual source code doesn't include changes from this build.
Quote from: Luzestudio on July 11, 2013, 11:16:06 AM
Read the first post: https://bitbucket.org/OtherOnePercent/tragic-lantern/
Quote from: marekk on July 11, 2013, 11:09:16 AM
When I can find source of this build ? I would like to port it to 60D..
Yes, please! 60D Nation is behind you. 8)
Before, I used to shoot with cinestyle, with 0 in camera sharpening, then, in Adobe After Effects, I had to use the unsharp mask effect, but with values like 1-250 to get good enough sharpening, and it was pretty noisy on the footage afterwards, but I had a solution to that by using the H.264 encoder, it gets rid of most of the noise, but still, it was a less than satisfactory result in terms on noise. After shooting with +1 in camera sharpening, in After Effects I was required to use only values like 0.5-100 to get the same sharpening that I got before, but with way less noise, like -90%. So it's definitely the way to go.
You can still use 0 in camera sharpening on scenes that don't need to be too sharp, but use +1 if you are not sure.
CBR 3x, 0 in camera sharpening, AE unsharp mask (1-250), f/5.6:
CBR 3x, +1 in camera sharpening, AE unsharp mask (0.5-100), f/5.6:
They look the same, but the first one, had way more noise before being exported with AE and encoded again with H.264, as a result of the high values of the unsharpen mask.
And another tip, if you want to get rid of all the aliasing from the image, shoot it with f/10-22. I know that every lens has it's most sharp f number, for some is f4, for others is f5.6, 6.4, but above a certain f number, the aliasing and moire start to go away. At f/10 it starts do go away, at f/16 it really starts to go away, and at f/22 it's almost all gone.
CBR 1.5x, 0 in camera sharpening, AE unsharp mask (1-250), f/22
<--- way less aliasing compared to f/5.6:
ps: these videos were NOT made with tragic lantern with the advanced H.264 controls.
I Don't See Any Of These Settings In My ML...... Slice Control.... Gop 1....... Etc.... How Do I Get Them???? Which Build?????
.......... I Got It.... Never Mind.... Feeling Noobish!!! :/
Canon 550D please...THX
Will this be available again in the newer releases? Tragic 2.0?
Quote from: dslrrookie on July 13, 2013, 05:09:14 AM
Will this be available again in the newer releases? Tragic 2.0?
Yes.
Can we tell you when ?
No.
:)
Have anyone reached more then 160Mbps average ?
Quote from: AriLG on July 15, 2013, 09:01:39 PM
Have anyone reached more then 160Mbps average ?
Thats about what I'm getting 140 - 160 average with jghardings recommended settings on a 45 MB/s card. My 30 MB/s cards won't keep up and stop recording after about 5 seconds.
Hi there!
I´m new here but I´ve been searching info about H264 recording in canon cameras, it´s structure, GOP lenght, Level of the profile etc..
Then I found this fantastic thread!! 8).
I read you are talking about eos 600D and 60D only (correct me if I´m wrong...) but I was thinking about the eos 50d wich is that I have!
I´ve been trying H264 recording but I missed some parameters like GOP lenght wich is very important for the final quality.
Are these settings available for the 50d implemented H264 module?
Another thing I´ve noticed in my 50d h264 recorded clips, is that they seem to be L5.0 Main profile, not Baseline
as I read in other articles arround the net!
What´s is the "main" diference between the two versions?
Thanks in advance fellows!
Cheers...
Asier.
I have used those settings that are in the first post of this topic, with ISO 80 and Slice 87, but I get bitrates from 86 to 96 mbs averege, is that normal ? I thought the settings will produce a minimum of 140 mbs, what am I missing ? My card is a Sandisk 45 MB/s formatted in exFat.
@Asiertxu
Pretty sure you can search for this, here, lemme do it:
http://www.magiclantern.fm/forum/index.php?topic=6888.msg59957#msg59957
@Critical Point
My understanding, reading ~100 posts for every question I ask (or searching for the answer), is that mb/s is ISO dependent. To get mb/s THAT high, you need to up your ISO. And ExFat does NOT affect write speed: that just gets you past the 4GB limit.
I wonder how is better, to use ISO 80 and ~90-100 mbs, or something like ISO 400 and ~120-140 mbs ?
It depends on scene complexity.
Static scene (ie. with no movement AT ALL. not the camera nor moving objects) will not produce high bit rate. on the contrary (the minimum for that setting). shoot a static noise video (like this : http://www.youtube.com/watch?v=DH0BQtwEAsM - download it at 1080p) to simulate a maximum movement achievable.... and watch the bitrate go sky high... it may even average at 175Mb/s ;)
Today I was filming from a moving car with the setting from the first post, I never got higher bitrates than 104 mbs, but I was filming at ISO 80. The lowest averege bitrate was 78 and the highest was 104 mbs, in different clips of course.
Thanks ALOT Yuppa for your help!! :)
I´ll try that link you provided me and see what I get! ;)
Cheers fellow!
Asier.
Quote from: jgharding on July 07, 2013, 02:39:20 PM
Yes you can do audio, in that it does function at 25p but If you turn it on with these settings the buffer fills straight away, so I leave it off ;)
Check this:
http://www.magiclantern.fm/forum/index.php?topic=1565.msg28974#msg28974
stable 160 mbps video in 25p with audio recording (higher than 165 mbps is not possible because of SD controller limit ~21mBps)
Quote from: Rush on July 19, 2013, 12:04:42 AM
Check this:
http://www.magiclantern.fm/forum/index.php?topic=1565.msg28974#msg28974
stable 160 mbps video in 25p with audio recording (higher than 165 mbps is not possible because of SD controller limit ~21mBps)
THX for the Info!
What Version do you use? TragicLantern1 or Nightly?
Quote from: Rush on July 19, 2013, 12:04:42 AM(higher than 165 mbps is not possible because of SD controller limit ~21mBps)
Actually it's 21.875 (175Mbps) :)
Cool! If I can repeat that ill change the post, is there a full settings list for up to 160 with audio?
QuoteWhat Version do you use? TragicLantern1 or Nightly?
It is TL v1. It is sad, but TL v2 disables sound recording automatically, if GOP or flush rate is changed.
Quote from: jgharding on July 19, 2013, 09:17:53 AM
Cool! If I can repeat that ill change the post, is there a full settings list for up to 160 with audio?
TL v1, CBR 3.0x, 25p PAL mode (strange, but 24p and 30p don't work with audio), GOP not higher than 4 (i prefer GOP4), fast card like SanDisk 95mb/s.
There is no flush rate hacks in TL v1, but flush rate can be very useful - it will make possible to use GOP higher than 4 (default GOP12 for example, that is more effective than low GOP).
QuoteActually it's 21.875 (175Mbps) :)
May be... but I see that my buffer start to filling up if bitrate reaches 170 mbps. It can peak at 170-180 mbps, but if it will be sustained for some seconds - it will overflow buffer (Card R/W tests don't go higher than 21.5 mbps for me with SanDisk 95 mb/s)
Ah I suppose lengthening GOP balances between buffer use and card writing. I do like I-frame though, but I suppose GOP 3 is a good compromise...
Quote from: Rush on July 19, 2013, 10:05:58 AM
It is TL v1. It is sad, but TL v2 disables sound recording automatically, if GOP or flush rate is changed.
TL v1, CBR 3.0x, 25p PAL mode (strange, but 24p and 30p don't work with audio), GOP not higher than 4 (i prefer GOP4), fast card like SanDisk 95mb/s.
And what Settings do you suggest in the "Slice CBR" Menu?
1% or anyone similarly knowledgeable: Is there a build with this functionality available yet for the 550D? If so, could you point to a link. I read the entire thread, but there are only little hints that it might be found somewhere. In any case, thanks for answering! Cheers!
Quote from: daSilva on July 19, 2013, 07:35:56 PM
And what Settings do you suggest in the "Slice CBR" Menu?
I had provided settings for CBR mode. I don't use Slice mode, so there is no any changes (slice control should be turned off. set slice to 0)
Tomorrow my new fast card is arriving, so I will be able to start using high bitrates.
I think I'll prefer audio in the recordings so I think the Rush version will be the one I'll try.
If I'm not wrong I need TL 1 and set the recording at 25 fps, is that ok?
@jgharding, did you have time to test this new configuration? Does it look as good as yours or there is a compromise for getting audio?
Thanks!
I shot this entire video with the 600D and TL- V1. GOP 3 (or possibly GOP 1, I'll have to check), slice 87, DBlock both on -1, some clips with NTSC 24p audio off and some with PAL 25p audio on. Most of the clips stayed around 90-120 mbps but a few hit 160-170. VisionColor PS.
You can even bump up sharpening in-camera a couple notches and not get that "digital" look, this build holds detail very well in h264 and colors/shadows can be manipulated a bit more.
Hi! I just received my Sandisk 95 mbs and tried with this configuration from @jgharding :
IN SLICE CBR
Lock Slice: Disabled
Min BR: 130
Max BR: 160
Drop by 1: 140
Drop by 3: 145
Taper Rates: Enabled.
IN BITRATE MENU
Mode: CBR
DblockA: -6
Dblockb: -6
PicPC: 0
GOP: 3 (changed from 1 to 3 as @Rush suggests)
Bitrate Info: On (I find it interesting)
BuffWarnLevel: 70% (just in case)
fps: 25
I can get audio.
My buffer is always between 30% and 40%
Card exfat formatted.
For now it seems stable.
Excellent, ill amend the top post to mention that GOP3 allows audio in 25p
Too much dblocking will result in increased noise, try it on -1 or -2.
And audio works on PAL 25P, but not on NTSC.
Here is my Virgin Media Shorts entry, shot with these settings and VisionColor
http://www.virginmediashorts.co.uk/film/5002/messiah
Quote from: N/A on July 31, 2013, 05:52:44 PM
Too much dblocking will result in increased noise, try it on -1 or -2.
And audio works on PAL 25P, but not on NTSC.
Tried with that option, not checked the results on the screen yet but I can confirm it's stable as before.
Buffer 30%-40% all the time with sound.
Quote from: jgharding on July 31, 2013, 06:10:26 PM
Here is my Virgin Media Shorts entry, shot with these settings and VisionColor
http://www.virginmediashorts.co.uk/film/5002/messiah
Very well! I like the visuals and the concept, the end is punchy.
How did you record the part of the roof where the camera is following the actor? I like it too.
In the first part of the short, on the kitchen, did you use additional light facing the character or was it natural? The photography of the short is really amazing, congratulations!
Best,
Xermán
Many thanks! :) The roof shot was a cheap Hague steadicam. I tried to move with his walking rather than make it dead still.
The kitchen shot used two LED panels and window light balanced, the rest is natural. In glad you liked it, thanks so much for the compliments!
Sh*tt! Now, with the enhaced video configuration I have a red spot on the left bottom of each video. It may be a dead pixel on the sensor, but it does not appear with stock or Magic lantern configuration.
Any idea of how to remove it? It's small but pure red and disturbs the videos.
Help will be much appreciated!!
Best,
Xermán
Sounds like a dead pixel.. screenshot? I've had none on 600D, have one on 6D that went away since the shutter was changed over.
you can change the GOP Size in the video hack menu in the SIXTHIRTY Tragic Lantern Built.
Is it the same like that way with the old build ?
2.0 has gop/flush independent so one or the other or both.
Here are 3 captures from different videos. I agree, it looks like a dead pixel.
I don't see it when I turn to default ML, so it must mean that we get more detail from the sensor now, the image is less smushed.
(http://i.imgur.com/oJwBuxu.jpg)
Any idea of a good way of removing it? I would really apreciate some way of taking it out, it's quite obstrusive and spoils the videos a bit.
Edit - Tried with a version of Neatvideo but nothing, it's still there. :/
Best,
Xermán
Quote from: Luzestudio on August 03, 2013, 12:33:43 PM
Here are 3 captures from different videos. I agree, it looks like a dead pixel.
I don't see it when I turn to default ML, so it must mean that we get more detail from the sensor now, the image is less smushed.
(http://i.imgur.com/oJwBuxu.jpg)
Any idea of a good way of removing it? I would really apreciate some way of taking it out, it's quite obstrusive and spoils the videos a bit.
Edit - Tried with a version of Neatvideo but nothing, it's still there. :/
Best,
Xermán
Have you tried this ....
The defect correction is off so things like this come out. That is a dead pixel for sure. You can mask it in after effects. Its only one.
Quote from: dslrrookie on August 03, 2013, 03:22:17 PM
Have you tried this ....
Ei, that seem to do the trick!! Thanks a lot, with that technique I got rid of the dead pixel.
Best,
Xermán
When i use the Settings from the first post, the footage gets really noisy.
Is it only because of the dblock settings or are there more settings to improve?
What are your favorite Settings?
The raw gets really noisy too, doesn't it? More detail.. too bad sometimes its mostly noise.
Neat video has an easier time getting the noise out though. And make sure to expose to the right just to the point of clipping highlights.
I´m not complaining. Of course Neat Video is doing a good job. I will try to change the dblock settings and look what i can improve.
Any other advises what i can/should try?
@1%
I´ve never tried RAW so i don´t know if it is as noisy as the bumped up h.264. ;)
I'll give some different Deblock settings a go and mention findings in the edit of the first post. In general I don't mind a lot of noise as long as it's clean noise, not mushed into the picture. That way Neat Video has a lot to work with!
@Sam, I feel like you, in fact I was going to show some of my latest footage, here is a comparison:
a: without noise reduction: http://youtu.be/-P11twryY5s (youtube compression helps alot, https://www.dropbox.com/s/oz2gei9l0t55xvs/Sequence%2001_1-NOISE.mp4)
b: with noise reduction: http://youtu.be/PNk8vS89iLw
The video at full screen 1080p will show a lot of noise in the dark areas, but I think that noise comes because we are lifting those underexposed parts. They are recorded with DblockA/B of 1. And flaat 10.
I film with noise of 80, 160, 320, 640, etc. I read it is the best option we have. By example, for iso 80:
Canon Analog ISO : 100
ML digital ISO: -0.3 EV
Equivalent ISO: 80
@jgharding, It's interesting what you say about "clean" noise, can you develop the idea a bit more?
Best,
Xermán
The Problem which i see with Neat Video and other noise reduction tools is,
that you lose sharpness. Shouldn't it be the goal to achive the cleanest, detailed
Picture we can get?
Bump up the sharpness in camera to 2 or 3 with dblock around -1. The picture style is applied to footage before the 8 bit encoding to h264, at the 12 bit level I believe, so in theory we would be sharpening 12 bit data in camera as opposed to 8 bit data in our NLE. Here's some more info about this.....
http://www.magiclantern.fm/forum/index.php?topic=4830.msg28671#msg28671
Then again, if you're shooting something with with horrible moire, you'd be better off to shoot with no sharpening in cam and sharpen in post.
Quote from: N/A on August 05, 2013, 09:01:25 AM
Bump up the sharpness in camera to 2 or 3 with dblock around -1. The picture style is applied to footage before the 8 bit encoding to h264, at the 12 bit level I believe, so in theory we would be sharpening 12 bit data in camera as opposed to 8 bit data in our NLE. Here's some more info about this.....
http://www.magiclantern.fm/forum/index.php?topic=4830.msg28671#msg28671
Then again, if you're shooting something with with horrible moire, you'd be better off to shoot with no sharpening in cam and sharpen in post.
Thanks for the link, interesting information.
Then HTP does something for improving the picture quality? I was totally wrong about this then...
For my likings i don't get good Results with any Settings.
Most of the Time it's just too noisy and finally you lose
detail in Post-Production.
@N/A
Thanks for the link. Really interessting. Camera Sharpening
may help, but if you have bad luck in some Situations you
and up with oversharped Footage.
Hi all,
First, thanks a lot for working on the EOS cameras :) I have installed the 2.3 ml on my 600 and it's working great.
I thought that it already came with RAW video but as I understand it having browsed around,
that it's not stable yet, hardly in HD and definitely not stable enough to shoot a short film with it.
So my questions are:
Is tragic lantern part of your team, or others, but in the end, are they complementary to your work or rival?
did you develop with tragic lantern a better control of the bit rate, to make better videos in spite of not having the raw stable yet?
And also, I'm lost with the different files they offer on the website: the six thirty, the new system and what you offer in this thread the bitrateback.
Lastly, sorry for being annoying, if i want to install these, do i install them on the sd card already containing magic lantern or this should go on a separate SD? And shall i just forget about the new system and just stick to the better bit rate?
Thank you :D
Is basically my repo. Code flows back and forth. It just frees me up to put 600D only stuff without worrying about breaking other cameras.
i.e. Movie mode remap isn't good for 60D but it works on 600D so it won't be in unified.
Do you plan returning all the GOP/Slice parameters to some newest build?
Improved sharpness
Due the high quality achieved with those settings I'm able to apply kernel convultion sharpening to the footable without visible artifacts and that gives, for my eye, a great sharpness, highly improved from my previous footage.
http://www.youtube.com/watch?v=pchhGTB3VvI
It's an improvised shoot w/o tripod. Tell me what you think about sharpness.
Best
Xermán
Quote from: Luzestudio on August 17, 2013, 02:53:34 PM
Improved sharpness
Due the high quality achieved with those settings I'm able to apply kernel convultion sharpening to the footable without visible artifacts and that gives, for my eye, a great sharpness, highly improved from my previous footage.
http://www.youtube.com/watch?v=pchhGTB3VvI
It's an improvised shoot w/o tripod. Tell me what you think about sharpness.
Best
Xermán
Looks Good! Where was this filmed?
The surface of the water looks great!
Maybe it's sharper that usual, but... still looks 720p. You can judge this from the logo in the end, which is supersharp (compared to the footage).
To make a comparison, the same exact video in 2 versions:
without convulsion sharpening: http://www.youtube.com/watch?v=qSX7FmWqBPw
with: http://www.youtube.com/watch?v=pchhGTB3VvI
I see them very different, the second one sharper, at 1080 it's very noticeable.
Best,
Xermán
ps - It's in Galicia, Spain.
Quote from: jgharding on July 04, 2013, 06:19:57 PM
This post is designed to answer questions about the H264 codec control in one version of Tragic Lantern in a simple and practical way.
After a few posts I made about using bitrate control on the 600D I had a bunch of messages about settings, and development threads were getting busy, so here's a new thread.
Though one could explain these concepts in greater detail, it's of no real practical use to the average user, so here they in plain English.
They are simplified, so lease don't fill the thread with highly technical responses on video encoding, that information is available elsewhere for those who are curious.
WHAT IS IT?
A balancing act of sorts. High bitrate means nicer footage but bigger files. You can adjust the colours more and remove noise more easily. Advanced control of the H264 encoder that tries to make it use as high a bitrate as possible, without fluctuating wildly or stopping recording.
WHY NOT USE RAW?
Raw can be lovely but there are also a lot of compromises (crop sensor, lower resolution, longer workflow) so perfecting H264 is a great way to get higher quality footage from your beloved camera.
H264 is limited because the encoder uses a rather soft image, but these settings will let you get the best from that image.
WHAT DO I NEED?
A 600D, a fast SD card of 45 or 95MBs will guarantee speed, and (currently) this build of 1%'s Tragic Lantern:
https://bitbucket.org/OtherOnePercent/tragic-lantern/downloads/autoexec.bin-bitrateback
The build is a little older than ones you may be used to so a few things are missing or different. GOP and slice control aren't in the new builds yet but may be at some point, and when they are let me know and alter this thread.
Some other useful things for the 600D are:
The Mosaic Engineering TXi filter, which removes most aliasing and moire, allowing you to make better use of the bitrate. Sometimes it gives Err30 butmost of the time it's amazing. You will still get some moire with exceedingly fine cloth and so on.
The Technicolor Cinestyle Picture Profile and it's LUT, which allows a flatter picture for grading. It's free.
The VisionColor picture profile, which is quite flat but has a nice natural film like tone. It's $6.99. That's a bargain, don't pirate it!
Neat Video denoising plug-in, which can make very high ISOs like 6400 usable when combined with the higher bitrates.
;D 8) THE SETTINGS 8) ;D
SOUND OFF *For maximum image quality, don't bother recording sound in camera, deactivate it in Canon menus. Use separate sound.*
IF YOU NEED SOUND use 25p mode and set GOP to 3 instead of 1.
IN SLICE CBR
Lock Slice: Disabled
Min BR: 130
Max BR: 160
Drop by 1: 140
Drop by 3: 145
Taper Rates: Enabled.
IN BITRATE MENU
Mode: CBR
DblockA: -6
Dblockb: -6
PicPC: 0
GOP: 1
Bitrate Info: On (I find it interesting)
BuffWarnLevel: 70% (just in case)
BUT WHAT DOES IT MEAN?!!
If you don't need to know, it's probably best to to worry. Hey, the important thing is it works! But here's some simplified explanations:
SLICE
Quality. The numbers used are weird, but not that confusing. The lowest numbers (86, 87) try and use as many bits as possible for encoding. The trouble is, if you lock the slice to 87 and give the camera a hugely complex picture the bitrate goes too high, the buffer fills up and recording stops, or quality has to drop for a while. The second of these situation makes the bitrate "bounce" up and down between loads and 40 or so, which means your image goes pixelly every so often.
Slice control lets the camera raise the slice (drop the bitrate) gradually and in a controlled manner if things get too much.
The settings I've provided here have been stress tested very heavily. They keep things at 87 a lot of the time, but reign it in if the scene becomes too complex. The settings are designed to keep the bitrate below 160 but as high as the camera can keep it without overflowing the buffer in complex scenes.
You can push it harder, but it won't be as stable. These setting have, for me so far, been the most consistant. I recently shot for 15 hours on a fashion video and never had any buffer drops. The average bitrate of the footage was around 100mbps I-frame, which is a lot better than stock.
The buffer can still get too full (resulting in a drop to a very low bitrate and some pixellation) if you shoot a very complex high ISO scene with the slice starting at 87 or so. If you are shooting a complex scene start the slice at 115 or so using the first setting in Slice Control, and the camera will quickly level itself.
GOP
GOP is "group of pictures". To compress more video into less space, some codecs look at one frame, then compare a lot of subsequent frames to it, only storing the bits that have changed. The GOP number is how many are in this group.
So what's the practical result? A long GOP is more efficient, but to my eye tend to look more video like, motion can be a little smeared. I prefer GOP 1, also known as I-frame, as each picture is a whole picture, like a piece of film. The downside here is that file sizes are big, but hey, it looks the best and memory cheap.
GOP 3 is a good compromise and allows sound recording in 25p mode with these settings.
BUFFER
A small memory where the camera stores things before they are written to the SD card. If this fills up, recording will stop. You can use GOP 1 to let frames be written straight to the card, but for some reason GOP 3 seems less hard on buffer. This is reaching the limit of my knowledge. GOP 1 probably stresses the card a lot, but your card will probably be obsolete before it matters. Just format it ater shoots.
Hi, as other users suggested before a Dblock of 1 gives less noise.
Best,
Xermán
Yes you can get away with more sharpening,
I like the sharpening in Neat Video due to frequency and channel specific...
What's kernel convolution sharpening mean?
I still need to test that deblock...
Quote from: jgharding on August 20, 2013, 04:39:24 PM
Yes you can get away with more sharpening,
I like the sharpening in Neat Video due to frequency and channel specific...
What's kernel convolution sharpening mean?
I still need to test that deblock...
Convolution Kernel effect
The Convolution Kernel effect changes the brightness values of each pixel in the clip according to a mathematical operation known as a convolution. A convolution overlays a matrix of numbers onto a matrix of pixels, multiplies each underlying pixel's value by the number that overlays it, and replaces the central pixel's value with the sum of all of these multiplications. This is performed for each pixel in the image.
The Convolution Kernel Settings include a set of controls that represent cells in a 3x3 grid of pixel brightness multipliers. Labels on the controls, which begin with the letter "M," indicate their position in the matrix. The M11 control, for example, affects the cell in the first row and first column of the grid; the M32 control affects the cell in the third row and second column. The pixel being evaluated falls in the center of the grid, at the M22 location. Use this effect for fine control over the properties of various emboss, blur, and sharpen effects. For a given effect, it is easier to apply one of the Convolution Kernel presets and to modify it, than to create the effect from scratch using the Convolution Kernel effect itself.
http://help.adobe.com/en_US/premierepro/cs/using/images/ae_ConvKernGrid.png
Convolution Kernel pixel grid, showing the position of each control in the matrix
from http://help.adobe.com/en_US/premierepro/cs/using/WS1E7C690B-2342-43c3-9253-2BAAFF4168EF.html
If you are into maths: http://homepages.inf.ed.ac.uk/rbf/HIPR2/convolve.htm
I read about it on this forum, don't remember where, but an user was recommending it. I like the results a lot.
I just tried Convolution Kernel but I really can't understand what's going on enough to tweak it in a way that is beneficial...
I put the preset in and tried changing things but I can't find a logic to it. It seems easy to make things ugly but I'm struggling to get the subtle results shown in the youtube video above.
I think it could do with a more friendly set of controls...
The sharpen preset is a bit excessive but I don't know how to reduce it...
For me nothing beats Neat Video yet. The luminance only, high-frequency only sharpening is great and almost artifact free. add a bit of mid frequency sometimes and you're good to go! easy and quick.
Quote from: jgharding on August 21, 2013, 01:39:59 PM
I just tried Convolution Kernel but I really can't understand what's going on enough to tweak it in a way that is beneficial...
I put the preset in and tried changing things but I can't find a logic to it. It seems easy to make things ugly but I'm struggling to get the subtle results shown in the youtube video above.
I think it could do with a more friendly set of controls...
The sharpen preset is a bit excessive but I don't know how to reduce it...
For me nothing beats Neat Video yet. The luminance only, high-frequency only sharpening is great and almost artifact free. add a bit of mid frequency sometimes and you're good to go! easy and quick.
Hi, Im using the default sharpen preset, it looks better with the lower Dblock (I think that makes the video "less blocky" in a way) and also everything becomes more natural when the video is exported to h.264
Best!
Xermán
I once saw the idea of changing the compression from h264 to mjpeg or something.
So we could get 4:2:2 or even better.
Is there still anyone working on that? maybe 1%?
RAW doesn't look like a proper solution on the smaller cameras like 600D
Quote from: Luzestudio on August 21, 2013, 04:39:56 PM
Hi, Im using the default sharpen preset, it looks better with the lower Dblock (I think that makes the video "less blocky" in a way) and also everything becomes more natural when the video is exported to h.264
Best!
Xermán
Cool I'll give it a try! :) I was trying on some stock 5D ii footage
Quote from: J_C_B on August 22, 2013, 12:43:16 PM
RAW doesn't look like a proper solution on the smaller cameras like 600D
This topic is not about RAW but about getting the most out of the existing compression.
This build is my favourite build, but it doesnt allow for modules. Would be awesome if this was combined with the other builds to allow for some of the other stuff, dual iso especially.
That way we wont have to switch builds for different purposes.
hello again, can anyone explain me how to install the GOP and Slice control on my 600D? I currently have the ML 2.3 running on the camera and the bitrate running at 3* but I would like to know how to go beyond that (my card is a lexar pro 95mo/s). Thanks
Quote from: Bosnik on September 09, 2013, 10:13:48 PM
hello again, can anyone explain me how to install the GOP and Slice control on my 600D? I currently have the ML 2.3 running on the camera and the bitrate running at 3* but I would like to know how to go beyond that (my card is a lexar pro 95mo/s). Thanks
See the first post in this thread.
download and replace your current autoexec.bin from link in post.
Thanks ! ;)
I didn't work at first because the filename is autoexec.bitrate and the original file is autoexec but it worked when I renamed it
I edited the original post slightly to reflect the findings of more testers, rather than just my own
Thanks for all the hard work! Great quality with these settings.
Hi, first of all I want to thank you all for the amazing job you have done with Magic Lantern! It's amazing how the filmmaking world has changed these last few years, professional filmmaking has become affordable and it's also thanks to you!
These settings are great I use them on my 600d with vision tech and sandisk extrem and it's super stable it gives just what you need for a more advanced color grading.
I've got one issue though, the playback on the camera doesn't work anymore. It's greenish laggy and it shows only one second then stop. If I don't record the sound it plays but still with weird artifacts and laggy images. Am I the only one to have that I don't see any post about it? Does anyone have a solution?
Thanks
The player can't play high rate videos.
Thanks for your quick answer. That's bad, how do you do when you want to check quickly your shot on location?
copy it or just look at what you can.
hi guys, amazing job!
Noob question: Is it possible to make a module that can edit these settings? With it, we could use newer versions of ML or TL without losing these features.
Thanks in advance, and sorry about poor english speech.
@jetrotal - exactly!
BTW, I think if it is possible to make available changing recording resolutions for H.264. If we could get higher buffer/without sound/2.39 widescreen, maybe final quality could be somparable to RAW. What do you guys think?
Quote from: AvdS on February 06, 2014, 01:06:10 AM
Thanks for your quick answer. That's bad, how do you do when you want to check quickly your shot on location?
You can play some of them back, other times you can skip forward to wherever in you clip and play maybe a few seconds from there.