Not to be overshadowed by Batch, Azure Blob Storage supports very rapid rates of access to those huge amounts of unstructured video and related data, making it the perfect choice to service VRT processing tasks and the corresponding variety of video and image types. Cloud9 has found itself at the leading-edge when it comes to using technology to help keep its teams ahead of the competition. And it’s working; the team has won four straight VCT Game Changers events.
The business of esports
For all sorts of businesses, AI and machine learning bring groundbreaking solutions to otherwise insoluble problems. But they also seem to bring an insatiable appetite for IT hardware performance and storage resources, a downside that can put the investment in AI beyond the reach of many. However, with the development of cloud infrastructure like Microsoft Azure, organizations have converted capital investment in those expensive resources—from hardware under desks to entire datacenters—into more manageable operational costs delivered by online services. With Azure, they can access not only the specialized AI and machine learning resources but also the virtualized, on-demand computing and highly performant, virtually unlimited cloud-based storage infrastructure required to develop and run those solutions.
That’s why esports company Cloud9 turned to Azure when it looked to provide its teams with a new, state-of-the-art competitive advantage built on AI and machine learning.
“One of our most notable projects is the Video Review Tool for the Valorant game. It was developed in collaboration with Microsoft and SOUTHWORKS, using Azure infrastructure, to provide our Cloud9 White team with new insights and a distinct competitive advantage that was previously unavailable.”
Halee Mason, Lead Data Scientist, Cloud9
Playing with Azure infrastructure
Cloud9 takes a comprehensive, holistic approach to supporting its teams, providing personal trainers, diet and nutrition advisors, and sports psychologists to help players maintain peak performance. Lead Data Scientist Halee Mason and her colleagues in IT offer their own unique contribution to Cloud9 teams’ continued success. “One of our most notable projects is the Video Review Tool for the Valorant game,” Mason says. “It was developed in collaboration with Microsoft and SOUTHWORKS, using Azure infrastructure, to provide our Cloud9 White team with new insights and a distinct competitive advantage that was previously unavailable.”
The Video Review Tool (VRT) began as a demo at a Microsoft Hackathon in 2020, at which time Mason recognized the potential of the technology. “We took the idea and built an entire data processing pipeline in Azure to run computer vision models against gameplay video and generate new data points and insights into the action,” she says. The VRT replaces the previously tedious and time-consuming process of manually searching footage of previously played games for teachable moments that coaches can subsequently highlight for their players.
The tool uses machine learning to train models against hours of video and identifies many more such moments than Cloud9 could manually identify in the past. Results are presented in a highly accessible, modern website for rapid online review and analysis. But it requires massive amounts of computing power, and correspondingly huge amounts of dynamic storage for the in-game video data that fuels the solution. So, a cloud-based build and deployment infrastructure was the optimal choice for Mason. “Cloud9 felt confident moving forward with Azure. After reviewing what we needed for this project, we saw that Azure already had all of the infrastructure services and components in place to successfully build the tool from end to end,” she says. “It readily supports deployment, video processing, and storage—all on demand, immediately scalable up or down.”
Compute power and efficiency from Azure Batch processing
Specifically, says Mason, Azure Batch and Azure Blob Storage provide infrastructure processing and storage. Mason says, “We needed Batch because VRT jobs are very long running. We have a video analyzer pool that is always pre-warmed and waiting for new jobs to queue, and the Azure Batch job is where all the magic happens.” The process begins with an ingestion script that uploads and stores the raw video in a storage account, then a message queue kicks off those magical jobs that engage the compute resources required to run the processing functions from an Azure Virtual Machines pool. Azure Batch automatically scales and schedules those resources according to the task at hand. Python scripts split video into single-second frames, on which computer vision models perform processing to classify key moments. Post-processing writes those events to a database containing time-stamped video, and the front-end service is updated to alert coaches and present the latest information in the VRT website.
The Azure Batch pool is where all that compute capacity is managed, helping to wring the most efficiency and performance out of available resources. “Azure Batch specifies the nodes required, and processing scripts run in containers managed by Docker. All of this is configured inside the pool level. Batch is a really critical part of this process,” Mason says. “There were major pain points involving these long-running jobs. Azure Batch addresses them, it supports and controls managing the video uploading, providing rapid cropping, and scaling through the container-based system.”
Multi-format data storage at scale from Azure Blob Storage
Not to be overshadowed by Batch, Azure Blob Storage supports very rapid rates of access to those huge amounts of unstructured video and related data, making it the perfect choice to service VRT processing tasks and the corresponding variety of video and image types.
Game video footage and cropped frames occupy most of the assigned capacity, Mason says, but she adds, “We have many data types and formats and Azure Blob Storage handles all of them.” The VRT currently uses some 225 gigabytes (GB) of production data, and there is another 190 GB or so in the development environment, along with a massive, and growing, amount of archived footage, “JSON data for every day describing hundreds of games, going back three years now,” Mason adds. Azure Blob Storage also offers geo-redundancy, which provides additional access continuity; “Geo-zone-redundant storage copies our data synchronously across three different availability zones and provides that backup and resilience we need,” says Mason.
Staying ahead of the competition
Cloud9 has found itself at the leading-edge when it comes to using technology to help keep its teams ahead of the competition. And it’s working; the team has won four straight VCT Game Changers events. Mason says, “The VRT we developed is just one example of how we've expanded our capabilities. Our tools, and Microsoft Azure, provide unique advantage. We're already scoping out additional opportunities to integrate technology and data insights to help our players perform even better in the future.”
And how does Mason see the future for Cloud9? “As the coaches use the tool, they’re excited about what we've provided and they bring new ideas, so the future is really bright and exciting for our contribution as data scientists to the continued success of team White and Cloud9!”
Find out more about Cloud9 on Twitter, Facebook, LinkedIn, and YouTube.
“We needed Batch because VRT jobs are very long running. We have a video analyzer pool that is always pre-warmed and waiting for new jobs to queue, and the Azure Batch job is where all the magic happens.”
Halee Mason, Lead Data Scientist, Cloud9
Follow Microsoft