Originally posted by AJ Jefferies
View Post
We were wary for the same reasons but the plan was if the i7's gave us trouble we'd use the ec2 cloud to fill in the gaps.
Martin -
It was done using this technique - https://www.youtube.com/watch?v=USS84b1iUp4
theres a full screenshot of this fusion file above, but this screenshot shows the footage projected back onto a 3d model, then how we 'rendered' the footage from a new camera angle which covered the area that needed painting/cleaning up. the cleanup was mostly simple - track the flat footage to get a transform offset to counter any wobble, freeze a single frame from any point in the footage, mask it over. job done! The more complex ones needed a few different frames frozen and mixed.
Then it's a case of rendering this new patch and nothing else from the main camera to overlay it back over the original footage. 11 of these later and we've got a clean plate!
I started working on some training material for this a while ago, it's a useful technique and I feel many people don't know it. it just takes too long to do properly though, we've kind of shelved the idea of a DBOX series of training materials, there's no guaranteed audience.
John O'Connell actually taught me a version of this technique based around rendering each pass from max 12~ years ago.
Comment