View Full Version : A question about workflows :)

03 March 2009, 01:19 AM
Hey guys & girls

I'm trying to develop a workflow, well more of a sort of pipeline structure to use for mixing live green screen footage (and non green screen footage) with CGI rendered shots.

My applications that i am using are Voodoo (although VooCAT is on the cards) and Blender.

I know there are many applications out there, but this is doing me just fine for everything at the moment.

Anyway my plan is/ was this;

Record live footage
Green screen footage with white markers for tracking, lit correctly to allow easier green screen extraction.

Edit and track live footage
Split video into TGA format, for frame editing and also for Voodoo (it only accepts TGA images).

Track live shot and tweak shot data, export to blender.

Extract green matte from image sequences in blender, using noodles.

Save out keyed frames for combining with CGI backgrounds (TGA + alpha files)
Render out CGI shots
Using camera with tracking data applied, render out CGI enviroments/ models, with all layers selected (alpha, Z depth, masking layers etc)

Render out as TGA format

Composite live footage and CGI elements in blender, with color correction and affects applied per layer (CGI layers and live layers, seperatley)

Render out shots for combining in timeline with sound effects and transitions

Basically that is my (roughly) planned out workflow and pipeline that i have, im just wondering if that will work, like am i missing anything major etc?

I would really appreicate feedback, thankyou :)

P.S. I have attached a quick screen shot of the keying section, i have had pretty decent results, however the film section i got from youtube to mess about with goes from green to a brown colour and im not sure how to key this out, how would the pro's do it? =/

03 March 2009, 04:30 AM
Firstly, wow, I didn't know Blender could do compositing! That's pretty cool.

In terms of suggestions, maybe try looking at red tracking markers (just red tape). Red on green tends to show up very well in the colour channels. Remember that things like tracking don't only looking at patterns (eg, an 'X' made with tape) but also at colour information like the luminance of the colour channels. White on green probably won't track as well as red on green.

Definitely try and light your greenscreen evenly and have actors stand well away from them to avoid shadows and spill. Shadows can really make a simpley key a lot more trouble than it should be.

If you are shooting DV and can get access to After Effects, Premiere, Vegas or Pinnacle Studio, I would highly recommend getting Neat Video noise reduction software. Keying DV footage sucks, but the results I have gotten out of Neat Video are stunning for the price. You run it as an effect on your raw green screen footage before outputting the TGA's.

Other than that, it looks pretty good. You might want to try a few test shots to see what render passes you will need for an average shot. Unless you have masses of hard drive space, it's just a waste to render every possible pass. If you need extra passes or mattes for a particular shot you can render them out on a shot by shot basis.

I have to say the more I look at Blender, the more impressive it is as an open source software project.

03 March 2009, 12:06 AM
DoubleSupercool: thank's for the feedback :)

It's much appreicatied, the gig i had planned however has been scraped, the guy wanted too much done for too little money. However i'm still taking in all that you said and i will get some green screen footage to test out. Also about the DV footage, would that be becuase of the filtering (or pixel?) ratio, like 4:1:1? Becuase i can split the channels from the images into YUV and just run a 2x2 blur on the U and V channels and come out with a decent image to chroma key.

Yea, blender has come some way since, well since last year really, if you ever feel like looking into it and have any questions, just shoot me PM :)

Thanks :)

03 March 2009, 05:07 AM
Hey Dan, sorry about the late reply. Re: the DV footage, yeah, the compression only samples the U and the V (chrominance) 1 time for every 4 on the colour. I believe you usually blur the U more than the V. That can definitely give better results for your keying, but it probably won't beat a dedicated noise reduction algorithm.

Seriously, this $50 Neat plugin was giving better results than the Furnace suite of plugins on the last feature we worked on. Reduced the noise in all channels but kept an amazing amount of detail on hair and edges.

03 March 2009, 01:34 AM
Looks like a well organized workflow.
I'm just a bit curious why you don't do the keying at the same time as you bring in the background/cg footage?
By rendering out with built in alpha, it would be a pain to correct issues in the key, and you would also want the background in there to see how well the edges mesh with it.

CGTalk Moderation
03 March 2009, 01:34 AM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.