Porto’s process involved doing a difference matte between his best manual solution and the computer’s own attempt. This internal battle of self-improvement is key to GANs success, the ‘counterfeiter’ keeps improving until it can fool the ‘cop’. One produces a result much like a counterfeiter, and the other tries to detect the fake, much like a cop. Internally, a GAN runs on the principle of two AI engines battling it out.
The solution was in how ML Generative Adversarial Networks (GANs) work. Once he had trained the ML engine, he started using Flame to generate ML ‘Thiago’ level marker removal.
#Autodesk flame nas software
He then got the software to train on his own before and after digital cleanups. Rather than using published data sets from the ML community, he gathered the best paint out shots he himself had done over his career.
Starting on Paint fixesĪ great example of Porto’s approach is how he started looking at the problem of painting out small faults such as rig removal or removing facial tracking markers. He believes the path forward is combining the science of ML research with the art experience of VFX professionals, and that such a marriage will dramatically alter effects work.
#Autodesk flame nas series
But Porto set out to see if he could incorporate a series of published academic ML algorithms into a predominantly 2D compositing pipeline. Porto will be presenting some of this online at next week’s online RTC conference June 9th.Ĭentral to Porto’s attitude is that experienced visual effects artists know an awful lot about making an image look right, and surely this wealth of experience, that ML researchers don’t have, is valuable. ML has become a key area of 3D R&D at most of the major VFX facilities. Porto set out to see if he could use the most recent cutting-edge ML in his beloved Autodesk Flame.
He’s interested in how developments in Machine Learning (ML) could be applied in visual effects and has been promoting ML and asking why there’s not been a lot more discussion on adding much more complex ML tools to visual effects compositing pipelines. Thiago Porto is a VFX Supervisor and Senior Comp at MPC New York. See the demo at NAB next week running on Amazon Web Services (AWS), with remote artist access from HP Teradici (HP Anyware), integrated by TrackIt, in the Hammerspace booth #N3043.Have you ever wondered about taking advanced new high-end Machine Learning approaches and applying them to regular VFX projects? For that matter, have you ever wanted to just be young again? “We are committed to extending and enhancing our cloud-enabled workflows to provide better solutions that support efficient collaboration at scale, which is why we are very excited to be at #nab2023 to demonstrate how Flame artists can leverage Hammerspace’s optimized data storage solution for collaborative finishing across multiple storage locations from virtually anywhere.” “Hammerspace’s qualification further enables our customers to move their productions to the cloud with a higher degree of flexibility around data and essence management,” says Steve McNeill, Director of Engineering at Autodesk. Big News! Scale Autodesk Flame workloads seamlessly in multiple storage and cloud locations with Hammerspace.
0 kommentar(er)
