First 7D alpha released!

Started by g3gg0, October 12, 2012, 10:36:53 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

jmalmsten

at what point does the bitrate become redundant considering that what we have to work with is 4:2:0 8bit?

3x8bit=24bit per pixel
1920x1080 is 2 073 600 pixels


2 073 600 x 24 = 49 766 400 bits per frame


49 766 400 x 24 fps = 1 194 393 600 bits per second... roughly 1 gigabit per second when in totally uncompressed mode. So with 100Mbps we get around 1:10 compression compared to the 1:20 that the standard can max out to in 50Mbps...

If my math is correct, that is, of course.

But at what point does the data start being redundant? I've seen some cameras hack out to 300Mbps. how much difference is there?

I mean.. like when they scan film... at some point, the resolution is redundant and all you're doing is recording grain-structure. Sure, I can scan Super8 in 8K... But it won't look that much better than a 720p scan since the grain is so big... for example... Or is it that it doesn't truly become redundant until we record 1080p with a 1Gbps codec? :)
My semi-bilingual site - >http://www.jmalmsten.com

g3gg0

now hit the limit...

...of my CF card i guess....

Help us with datasheets - Help us with register dumps
magic lantern: 1Magic9991E1eWbGvrsx186GovYCXFbppY, server expenses: [email protected]
ONLY donate for things we have done, not for things you expect!

g3gg0

someone here with a card benchmark that shows more than 33MiB/s ?
Help us with datasheets - Help us with register dumps
magic lantern: 1Magic9991E1eWbGvrsx186GovYCXFbppY, server expenses: [email protected]
ONLY donate for things we have done, not for things you expect!

szigiszmund

hi , my sandisk card shows writing buffer=128k :26.9,end writing buffer=16384: 60 mib/s

ilguercio

Quote from: g3gg0 on October 25, 2012, 12:22:27 AM
now hit the limit...

...of my CF card i guess....


Wow, but is it worth the epic size?
Canon EOS 6D, 60D, 50D.
Sigma 70-200 EX OS HSM, Sigma 70-200 Apo EX HSM, Samyang 14 2.8, Samyang 35 1.4, Samyang 85 1.4.
Proud supporter of Magic Lantern.

feureau

Quote from: g3gg0 on October 25, 2012, 12:53:25 AM
someone here with a card benchmark that shows more than 33MiB/s ?

I was just about to go out and get a new card. How fast do you need?

I can try some cards out at the store.

feureau

Quote from: ilguercio on October 25, 2012, 01:37:33 AM
Wow, but is it worth the epic size?

If it is possible to mitigate the 4gb limit and shoot continuously after hitting 4gb: yes! If it's not, well, we'll just have to shoot shorter scenes, so: YES!

P337

Quote from: feureau on October 25, 2012, 05:11:22 AM
If it is possible to mitigate the 4gb limit and shoot continuously after hitting 4gb: yes! If it's not, well, we'll just have to shoot shorter scenes, so: YES!

Well at about 245 Mbps that's like 2 minutes of footage per 4GBs (if it's a constant bitrate that is) but that should be enough for one scene :)  and we would need at least a 30MB/s write speed so something like a "500x" 40MB/s Card would be ideal.  I've seen them sold for about $2.50 per gig so you could think of it as $5 per minute ;D

Any chance of seeing a resolution chart video with the 7D at 200mbps :-D

mimiloveyou

The GOP is it changed?

Very good job.
Thank you.

P337

Quote from: mimiloveyou on October 25, 2012, 06:35:14 AM
The GOP is it changed?

Very good job.
Thank you.

I assume it's still IPP, which (correct me if I'm wrong) is really only a concern for recording fast erratic movements.  Increasing bitrate helps resolve higher levels of detail in the image which I think is more important. (Though I've also heard the 5D3's "ALL-I mode" helps reduce morie and aliasing but haven't seen for myself)

I remember reading somewhere that in Canon's encoding process, 1 out of 15 frames are fully recorded and the rest attempt to "predict" the frame; in the "predicted frames" only the changes from the previous frame are recorded and anything that did not change is basically copied and pasted to the next frame so adding "B-Frames"(IPB) to that mix would be very helpful for those "Predicted frames" but I think that would require messing with Canon's code which ML doesn't do.  Making it Intra however might be easier by telling the encoding "OMG every single pixel changed since the last frame, so quick record it all!" for every frame lol.

But these are all guesses on my part, I'm not an engineer just a observant user ;D

mimiloveyou



P337

Quote from: mimiloveyou on October 25, 2012, 08:13:57 AM
GOP 3 with EOS 600D

http://vimeo.com/49988361

Oh cool, I didn't realize ML already did GOP controls!  GOP3 is close enough to ALL I for me, hope this can port to the 7D! ;D

P337

Quote from: feureau on October 25, 2012, 08:49:51 AM
What means GOP 3?

It means only 2 frames are being predicted per "group of pictures" instead of 14 in the standard IPP compression.  "All-I" means no predicted frames so "All-I" is GOP1

Shizuka

It should be noted that for the same bitrate, shorter GOPs yield significantly poorer compression. GOP=18 is more efficient.

P337

Ah this guy shot with a ML beta 600D in All-I mode! And apparently I heard wrong; stock Canon encoding is GOP12 not GOP15. 24fps recordings are GOP12 while 30fps and 60fps recordings are GOP15

"Gettin' High - Super Neutral Log Picture Style Test" on Vimeo http://vimeo.com/andy600/gettinhigh

g3gg0

i didn't check GOP, but it is at its default value. maybe 12, maybe 18. not sure.

didnt play with that yet. i just increased the write rate by flushing buffers more often, so a higher bitrate wont fill the buffers.
result is a possible, constant card write process at maximum rate of CBR x9.0 (LED permanently lit)
tried with 600D too, but my SD card couldn't handle more than 150mbit/s.

i recorded my monitor which displayed 'cat /dev/urandom | hexdump ...' in fullscreen to get that high bitrate.

i will analyze the video content tonight. but i remember that canon only uses I and P frames.
they do not implement B-frames (bidirectionally predictive coded).
however P-frames are the ones containg a lot of information from previous frame if possible, mixed with
some new data, unique on the current frame.

the 4GiB limit is not likely to get removed in 7D, as it works with 32 bit file sizes.
600D works with 64 bits, so this one is possible to override 4GiB limit.
Help us with datasheets - Help us with register dumps
magic lantern: 1Magic9991E1eWbGvrsx186GovYCXFbppY, server expenses: [email protected]
ONLY donate for things we have done, not for things you expect!

g3gg0

Quote from: P337 on October 25, 2012, 09:27:22 AM
Ah this guy shot with a ML beta 600D in All-I mode! And apparently I heard wrong; stock Canon encoding is GOP12 not GOP17.

correct, default GOP is 12. but on 7D there are also encoder options that say 18. maybe thats 720, didnt check.
Help us with datasheets - Help us with register dumps
magic lantern: 1Magic9991E1eWbGvrsx186GovYCXFbppY, server expenses: [email protected]
ONLY donate for things we have done, not for things you expect!

P337

Quote from: Shizuka on October 25, 2012, 09:24:09 AM
It should be noted that for the same bitrate, shorter GOPs yield significantly poorer compression. GOP=18 is more efficient.

You mean requires higher compression rates.

Right, cause you're leaving less information headroom for each frame, that's why I think a higher bitrate is more important than changing GOP; a higher bitrate means less compression so you get fewer compression artifacts and more detail while shorter GOP means less predicted frames so you get more accurate fast movements and more natural motion blur.

Thanks for looking out Shizuka

mimiloveyou

A small GOP. is best quality.
Details in the image are thinner. (fine details)

Sorry for my bad English.

sanderbontje

I ran into two small things.

1) Does anyone else have trouble disabling the white graph that looks a bit like an exposure history graph? At least that's what I think it is, it's the little white graph that is refreshing slowly from the right and placed on the bottom right. Or is it a CPU load graph? This one:


I can reproduce it like this:
- a fresh and clean install of ML
- hit trash can, disable all overlays individually except magic zoom (and Global Draw is on, of course)
- hit Live View and the white graph is still there

- in Display > Advanced Settings change the screen mode to '16:10 HDMI', the graph is gone.
- same for all other screen modes except '4:3 Movie'
- change back to screen mode '4:3 Movie', and the white graph is back again.

2) I would have liked to include a few screenshots, but the screenshot action does not seem to work right for me. I selected the 'Screenshot - 10s' option in the Debug menu, pressed Set, browsed to the settings page I wanted to take a screenshot from and expected it to take a screenshot after 10 secs and save it on the CF card. After about two or three seconds the camera bleeped and the red 'writing to card' LED lit up for a moment. There was no blinking each second to indicate a countdown.

After a few attempts I found several screenshots on the card with file dates that correspond to the time of testing, but they are all named TEST.BMP (yes, each one). Next to them were some VRAMX.422 files with the liveview images. The bmp's all contain exactly the same screenshot taken from the Debug menu, like this:


The YUV 422 images all seem okay though, after converting them (using this converter).

Maybe I am doing something wrong, or is the screenshot mechanism not yet ready for the 7D? What is the recommended procedure?

RenatoPhoto

Will HDR Video be included in Alpha 2?  Need it soon..
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

StefanKeller.AC

Quote from: dutchguy on October 23, 2012, 10:26:50 AM
I noticed a problem when using ML: when I tried to focus normally (not using LiveView) using a focus point other than the middle focus point, the camera could not lock the focus. When I turned off the camera (so ML was deactivated) and tried again, focus was working fine.

I have not tried this again, so it could be a one-time problem, but maybe someone else can try to reproduce?


I had this two times! (in good light-conditions) 
After first restart AF worked for 2 or 3 pictured than failed again, center AF-Point worked, after restart it works with and without ML...

feureau

Quote from: StefanKeller.AC on October 25, 2012, 05:40:55 PM
I had this two times! (in good light-conditions) 
After first restart AF worked for 2 or 3 pictured than failed again, center AF-Point worked, after restart it works with and without ML...

That's interesting, I'ven't been having any of these issues. What focus mode have y'all been using? (i'm using AI Servo a lot so that could be it)

feureau

Quote from: sanderbontje on October 25, 2012, 11:57:50 AM
2) I would have liked to include a few screenshots, but the screenshot action does not seem to work right for me. I selected the 'Screenshot - 10s' option in the Debug menu, pressed Set, browsed to the settings page I wanted to take a screenshot from and expected it to take a screenshot after 10 secs and save it on the CF card. After about two or three seconds the camera bleeped and the red 'writing to card' LED lit up for a moment. There was no blinking each second to indicate a countdown.

After a few attempts I found several screenshots on the card with file dates that correspond to the time of testing, but they are all named TEST.BMP (yes, each one). Next to them were some VRAMX.422 files with the liveview images. The bmp's all contain exactly the same screenshot taken from the Debug menu, like this:


The YUV 422 images all seem okay though, after converting them (using this converter).

Maybe I am doing something wrong, or is the screenshot mechanism not yet ready for the 7D? What is the recommended procedure?


Screenshot feature delays screen shotting for 10 seconds. Just hit screenshot and navigate away to what you want to screenshoot. Wait until it beeps. Beeep means screen is shot.

Rename test.bmp before copying off the card. Yes, they're all called test.bmp but they're from different screen shots.