The REAL difference between normal DSLR video and 5D Mark III raw video

[Andrew Reid]

Stephen, what makes the most difference in your experience - bit depth of colour sampling? Is 8bit unfairly maligned? Because the accepted wisdom in the DSLR community is currently that 8bit is the poster child for poor image quality.

 

I've seen uncompressed 8bit 4:2:2 from the 5D Mark III's HDMI port and wasn't impressed. It looks like a similar result as you get with the internal compressed codec. But could we be pinning the blame on the wrong thing? The problem is, there's tons of data thrown away BEFORE the compression stage. Removing the compression makes no difference as by then the camera has already thrown away most of the sensor data. With raw it doesn't throw anything away after the feed leaves the sensor. The only loss is from the binning of resolution down from 22MP to 2MP on the sensor itself.

 

My impression was that the uncompressed 5D Mark III HDMI output in the latest firmware was still debayered very poorly and heavily processed in-camera before it reached the HDMI port. They claim 4:2:2 colour sampling. I don't think it is debayered to 4:2:2 in camera. I think the signal is 4:2:2 but the data is the same old shit.

 

The difference to the 8bit BBC broadcast stuff is that from sensor to media, the whole imaging pipeline is optimal. A bit more like the JPEG engine on a DSLR rather than video. Much less is thrown away.

 

So what matters is more the entire pipeline - sensor sampling, A/D conversion, signal processing, debayering, encoding, compression quality and media performance.

 

The problem with 8bit is with a heavy grade - no matter what the quality of the source - you are going to run out of gradation and tonal precision at some point, which risks introducing banding.

 

Laptop on set..

 

I recommend Macbook Air from 2012 onwards. Fast USB 3.0 to transfer the card data quickly. CPU is quick enough to handle the debayer and playback smoothly. The software side is improving on a month to month basis.

[Stephen de Vere]

I think it's a community thing - yes, 8-bit in the DSLR community is bound to have a 'bad' name compared with 8-bit in broadcast worlds.

 

I can't really say whether in my experience sampling regime or 8-bit/10-bit etc. makes the bigger difference. There are just too many other elements that play into the mix.

 

Yes, basically 8-bit gives you very little leeway, even when it's in a professional implementation like DVCProHD that eg. Planet Earth was shot with - you need to get the exposure right, in camera, very close to what you want in the final. In my professional experience (mainly BBC) that exposure accuracy is by far the major issue - because it's pretty much impossible to get it right every time in an outdoor, uncontrolled and often rushed filming environment.

 

The big 'killer' I think is the combo of 4:2:0 (eg. normal DSLR and consumer camcorders) AND savage compression ratios. Of course, as you say, the whole pipeline from sensor to recording media is critical too, but I'm pretty sure that 4:2:0 is a major cause for banding and blocking issues.

 

For my personal independent films - largely destined for DVD, BluRay, VOD/internet delivery - I still shoot 8-bit 4:2:2 140mbps I-Frame XDCAMHD (EX3+Nanoflash). It looks great most of the time partly because I've got 20+ years experience filming ! On the other hand my 2nd camera footage - 8-bit 4:2:0 high bitrate I-frame hacked GH2 - while often almost indistinguishable from the 1st cam, gets caught out by eg. banding in blue skies and long-lens blurry backgrounds.

 

I don't have experience with HDMI connections. I suspect you are right that just because the HDMI spec is 4:2:2 it's possibly misleading and you can only know what the actual signal/output is by checking it directly.

[Margaret David]

very funny video hahahha