Video quality - what is the next push?

Started by skydragon, November 05, 2012, 11:28:36 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

skydragon

General question for Alex and the development team.

What do you see as being the next stage in ML development to get better video quality out of the camera? By 'better' I mean higher quality 1080p video with less artifacts and more detail, in a fairly easy-to-use and understand video recording UI. (FWIW - for me, getting better video quality is my No1 wishlist item)

I've seen on the forum (but don't really understand) some All-I discussions, mention of MJPEG and also of (distant?) potential for clean 1080p HDMI out and external recorders.

Taking a snapshot of the current ML work and the current known EOS environment, what's next for better video quality?

(As always thanks guys for all the hard work)
Canon 600D with Battery Grip, ML 2.3, Sigma 17-50mm 2.8, Swivi viewfinder loupe, RJ Follow focus. Loupe, FF and Camera/Battery Grip mounted on a 4mm thick alloy plate.

1%

Resizing things to real full HD or recording mjpeg. The clean HDMI might be possible too.  I don't know about simple though.

I can record YUV silent pics at full size and that is supposed to be "best" quality but I'm getting like 2fps so not really doing much for video.

600D H264 quality already as good as its going to get. Only thing left is resizing YUV to be larger which may or may not be mathematically possible if you read the 7D thread. Also those "higher quality" things require 64bit addressing or you're limited to only 4GB files. Looks bad for most other cameras.

All I can say is we'll see where it goes.

skydragon

Thanks 1%

I wonder if >>or you're limited to only 4GB files<< is really a problem for most ML users, if it meant better video quality as an option ? (question not a statement)
Canon 600D with Battery Grip, ML 2.3, Sigma 17-50mm 2.8, Swivi viewfinder loupe, RJ Follow focus. Loupe, FF and Camera/Battery Grip mounted on a 4mm thick alloy plate.

Francis

Quote from: skydragon on November 05, 2012, 11:54:55 PM
Thanks 1%

I wonder if >>or you're limited to only 4GB files<< is really a problem for most ML users, if it meant better video quality as an option ? (question not a statement)

Better quality = larger files = shorter clips. This is definitely an issue for a lot of ML users.

g3gg0

the bitrate increase i experiment with on 7D is also possible (tested) on 600D.
but SD cards are not as fast as CF cards. at least mine is too slow to increase rate by a significant amount.

still this might be an option to produce better footage.
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!

skydragon

g3gg0 - regarding the
Quotebitrate increase i experiment with on 7D is also possible (tested) on 600D
can you explain further please? Are you referring to the existing ML 2.3 ability to crank the bitrate up by altering the CBR factor (eg. Taking CBR x1.0 to say CBR x 1.8 ) - or is this something else? I noticed another thread regarding your experiments in massively increasing bitrate, with CBR x9.0 mentioned....but... I got the impression from my limited understanding that just simply increasing the bitrate above say CBR x2.0 won't actually yield any real-world advantage in perceived video quality.  FWIW - I'm already running my 600d successfully at CBR x1.8 for all usage (with no audio) but struggle to see any difference if I crank my bitrate up to CBR x2.8.

If the resizing of YUV in the camera isn't technically possible, then I presume that the only avenue left is regarding MJPEG based recording or clean HDMI out to an external recorder?

Francis - Good feedback, I guess we all have different usages and expectations, which is no bad thing. I would hope that users who don't want to have to deal with very large video data files, can choose to use the normal bitrates and recording regimes.
Canon 600D with Battery Grip, ML 2.3, Sigma 17-50mm 2.8, Swivi viewfinder loupe, RJ Follow focus. Loupe, FF and Camera/Battery Grip mounted on a 4mm thick alloy plate.

Luiz Roberto dos Santos

Hello Guys! Any news about recording in MPEG? 1%?

hjfilmspeed

Interesting post over here: http://www.magiclantern.fm/forum/index.php?topic=3489.0

http://vimeo.com/52822486

im not sure if i would want to record my video this way yet. but i think this guy may have revealed why the mark 3 has such soft internal recording. 1620x910 is probably what the mark iii is actually resolving (if even). then the 8bit and 4:2:0 h.264 probably further cripple it. At this point the olpf thing is deffinetly out. Again shooting an s2 jpeg in video mode yields a 1920x1080 jpeg that looks way sharper then its video. (realy good actualy telling me the rez is there, just not at 24fps) The soft video of the mark iii is due to how its sampled in camera IMO. Weather canon deliberatly did this or the camera cant physically capture true 16X9 1080 from its 20+mpix 35mm sensor is still unclear i guess. I do remember canon saying they were "trying to preserve details." and that they expect mark iii users to step up there "sharpening skills" Im not trying to bash but that to me sounds like a line there supposed to say so we feel at fault. I wish canon would just be straight with us is all.

g3gg0

Quote from: skydragon on November 06, 2012, 10:56:49 AM
g3gg0 - regarding the  can you explain further please? Are you referring to the existing ML 2.3 ability to crank the bitrate up by altering the CBR factor (eg. Taking CBR x1.0 to say CBR x 1.8 ) - or is this something else? I noticed another thread regarding your experiments in massively increasing bitrate, with CBR x9.0 mentioned....but... I got the impression from my limited understanding that just simply increasing the bitrate above say CBR x2.0 won't actually yield any real-world advantage in perceived video quality.

its exactly what you explained. increased CBR to larger values.
tests have shown that the video quality increases especially in areas with many details and when moving camera.
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!

Luiz Roberto dos Santos

Thanks for your response hjfilmspeed! Any chance of recording such a 600D? ALL-I recording is already happening, I'm watching, but I say about recording in MJPEG?

3pointedit

I still don't have an issue with shorter clips if the quality is better. I rarely shoot anything longer than 5mins at a  time anyway.
550D on ML-roids

1%

That 5 minutes can be 5gb. For MJPEG the tethering has to be fixed

hjfilmspeed

id take it if it ment real 1080p and fine
film grain like noise of a jpeg

1%

Can't resize it yet and not sure about the frame rate. Needs more work but probably the only chance to squeeze any more quality out of these.

tickerguy

There's a problem here in that it APPEARS that Canon is reading every third line (or thereabouts) from the sensor.  This is why you're getting the moire patterns and it's also why it appears that horizontal and vertical resolution in video lines are not equal.

That may not be fixable; to get the full quality the sensor can resolve you have to be able to read the entire sensor at the frame rate (e.g. 30 fps) and interpolate in the firmware.  It does not appear that any of these cameras can actually do that physically -- nor do they have to in order to shoot stills.

If you can't clock the data out of the sensor fast enough then the rest doesn't matter as you don't have the raw sensor data to interpolate; you cheated when you read the sensor, which limits your ultimate resolution.

From what I can determine this is what they're doing and it makes sense; even the 7D, which can shoot 8fps stills, only need to be able to read the sensor in 1/10th of a second or so at full frame rate.  There's quite a difference between that and reading it at 30fps or more.

1%