Identifying potential areas where machine learning can have a direct impact on visuals is especially easy, not least because there is so much focus currently on analysis and synthesis of image data. But there are opportunities to streamline many other aspects of the visual effects pipeline, allowing more efficient use of finite resources (such as artist time, software licenses, or hardware). Dozens of decisions are made daily which aim to maximize throughput whilst keeping projects to schedule and under budget. Some of these might be purely technical, for example “which shots should be prioritized for rendering tonight?” or “how much storage will this show require?”, whereas others might be more abstract such as “what are the risks to this project?”. Multiple sources of data may need to be examined to infer reasonable answers to such questions.
Here’s a great example of machine learning being applied to production processes, originally presented by MPC as a talk at Siggraph 2017. The write-up, as you might expect from subject matter that deals with sensitive company data which may be of value to competitors, is a little light on specifics (the talk itself may have gone into slightly greater depth). In particular, no baseline for comparison is established making it harder to evaluate the success of their approach, but it does serve to demonstrate how companies are actively exploring the problem space. There will surely be more to come.
VFX production companies are currently challenged by the increasing complexity of visual effects shots combined with constant schedule demands. The ability to execute in an efficient and cost-effective manner requires extensive coordination between different sites, different departments, and different artists. This coordination demands data-intensive analysis of VFX workflows beyond standard project management practices and existing tools. In this paper, we propose a novel solution centered around a general evaluation data model and APIs that convert production data (job/scene/shot/schedule/task) to business intelligence insights enabling performance analytics and generation of data summarization for process controlling. These analytics provide an impact measuring framework for analyzing performance over time, with the introduction of new production technologies, and across separate jobs. Finally, we show how the historical production data can be used to create predictive analytics for the accurate forecasting of future VFX production process performance.
Are you aware of some research that warrants coverage here? Contact us or let us know in the comments section below!