TV Soup – or how video compression really works

A little while ago, I got embroiled in a discussion about editing footage from DSLRs and why it wasn’t always a good idea to desire editing the original camera files. I repeat a condensed version of rant here for some light relief – but please can you imagine it as delivered by the inimitable Samuel L. Jackson…

When your DSLR camera records video, it needs to be space efficient as it has to deal with a lot of frames every second. Merely recording every frame does not leave enough time to actually capture subsequent frames and compress them nicely. It needs to do some Ninja Chops to do video.

Firstly, it does not record each frame as an image. It records a frame, and for every subsequent frame it only records the changes from the first frame. This may go on for, oooh, 15 frames or so. Then it takes a breath and records a full frame, then does the differences from THAT frame onwards.

Now imagine you are an editing application. Scooting around in that framework of real and imaginary frames means you’re spending most of your time adding up on your fingers and toes just to work out which frame you’re supposed to be displaying, let alone uncompressing that frame to display it.

Oh yes. In order to edit, you have to DECOMPRESS frames to show them, and that takes time. It’s like making ‘packet soup’.

Your editing software is trying to snort up packet soup – dried bits of vegetable and stock – it has to add a specific amount of water to that mix, allow the dried bits of aforementioned stuff to absorb the water, then compartmentalise the soup into spoonfuls.

Lesser compressed soup (not H.264 freeze dried but ProRes/DNxHD ‘just add hot water’ concentrate) can do this quicker and better – and some say it tastes better too. If only these newfangled cameras stopped freeze-drying their soup and just stuck to boiling off the excess water like MPEG2 does, dang, that would be nicer.

So, when you take your camera originals in H.264, you have to carefully re-hydrate your freeze-dried movies, and allow them to slowly absorb their moisture in a long process called transcoding. Then gently simmer them to a stock soup concentrate, so your editi system can easily serve them up in 1-frame, 1-spoon servings so you can edit them between the many hundreds of thousands of bowls that maketh the feast of your film.

You can have QuickTime soup. You can have Cineform soup. You can have DNxHD soup. H.264 soup is freeze dried and acquired through a straw. But H.264 soup is the size of a stock cube, and (for want of a better example) R3D is like canned soup – just requires a little reheating and a cup of cream.

Which ever way you capture and store it, we all watch soup.

Take your T2i footage, rehydrate it into the editing format you choose (can be ProRes, DNxHD, Cineform, hell, even XDCAM-EX) and then dish it up by editing and add your secret sauce to make it look/taste even finer. When you try to edit raw footage on most edit systems, you’re making soup into a condiment.

Thank you Mr Jackson.

Okay already, enough of the metaphor (and you’re spared the spatial compression stuff for now). CS5 does the ‘edit native H.264’ trick very well, so can other systems in the future, no doubt. But there is most definitely a time and a place for transcoding before editing. And I don’t think it’s going away.

Advertisements