The visual effects industry is no stranger to disruption. It has been shown to be both remarkably resilient yet also occasionally fragile at the same time. There are a handful of large, successful companies which have been in existence for decades, yet others have failed to endure through more challenging times. Some have grown rapidly to become dominant players in relatively short periods of time, others have attained size by acquisition. However you look at it it is safe to say that turbulence is to be expected again at some point in the future. Machine Learning is often referred to as a disruptor. It has the power to bring about dramatic change in many industries in ways that might be surprising.
“We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten. Don’t let yourself be lulled into inaction.” – Bill Gates
We’ve previously written about the importance of having large quantities of quality training data. Historical data may not always be easy to find but the success of machine learning in VFX may depend on it. Are companies with well-established asset management systems best poised to succeed, or might their legacy toolsets hold them back? Conversely, would a new company unencumbered by legacy technology be at a disadvantage without a rich history of projects from which to mine data and, just as importantly, a diverse set of client relationships and projects to enable them to keep the lights on?
It may be foolish to assume that simply because a studio has delivered dozens of projects they’re now awash with rich data which can be readily accessed, interpreted or analyzed in ways which will significant impact their bottom line. Some of it might have been discarded – or never collected – rather than retained. Is it possible that an as-yet unknown entity is capable of developing a data-driven pipeline so efficient that they’re able to crank our shots for pennies on the dollar and dethrone the current kings? Could gigantic companies such as Apple and Amazon, who are themselves investing heavily in content creation, find themselves in an advantageous position based on their machine learning know-how which gives them the capability to perform VFX in-house and ultimately compete with – or even displace – the major studios?
Larger players might be financially able and willing to invest in R&D in pursuit of such technologies but the very nature of research makes the returns hard to quantify ahead of time. The risks are high, but so are the potential rewards. One pattern that remains a relative constant in VFX is that that which was proprietary state-of-the-art technology very quickly becomes a commodity as it is inevitably replicated and rolled into the feature set of standard 2D/3D applications. This may open the door to all, regardless of their investment. Competitive advantage may be short-lived as a result. So-called AutoML may act to level the playing field even further.
We might already be seeing a glimpse into the future as rapid advances in face replacement technology suggests how work which traditionally requires hundreds of hours of tracking, modelling, animation, look development, compositing and so on might soon get reduced to a few button presses after collection of required training data.
Will machine learning be VFX’s next big disruptor? Or is the hype simply not justified? Are the people more important than the technology? Let us know in the comments, we would love to hear your thoughts.