I Went to the NHK STRL Open House 2023
This article was automatically translated from Japanese by AI. Read the original Japanese version →
Table of Contents
On Sunday, June 4, I went to the NHK STRL Open House.
You can check out very detailed information about the exhibits at the link below.
Open House 2023 "Supporting Media, Creating the Future"
In this post, I'll jot down my impressions and personal takeaways.
Volumetric (3D Video Capture) Exhibits
One exhibit that left an impression in 2022 was "3D Information Acquisition via Meta Studio."
3D Information Acquisition via Meta Studio
Roughly speaking, the idea is to build a capture studio that surrounds the person or object you want to film 360 degrees with cameras, turning the subject into 3D video data. The advantage is that you can freely decide camera angles during editing—or, depending on how it's delivered, let viewers move their viewpoint around freely. If you've ever experienced a VR concert, it's a similar concept, except realized with live-action footage.
So what updates did 2023 bring? Roughly speaking, there were two main ones.
3D Spatial Authoring Tool
Even after creating 3D video data, turning it into proper content for distribution requires dedicated editing software. Think placing objects in 3D space, setting boundaries on where users can move, defining default viewpoints, and the like. Since some people will experience the content in VR while others will watch it as standard video, they seem to be building it to handle both in one unified workflow.
Automatic Camera Viewpoint Selection Technology in 3D Space
Automatic Camera Viewpoint Selection Technology in 3D Space
Having humans manually pick the best camera angles when viewing a 3D model is a lot of work, so the idea here is to let a machine do it. You'll get a better picture from the link, yet the approach itself—computing good camera angles algorithmically—struck me as pretty unique and interesting.
AI Exhibits
Accessibility Support Technology
Automatic Program Video Summarization System Using Image Analysis AI
For someone like me who follows AI news on a daily basis, honestly there wasn't much here that felt brand new, though it's clear NHK is actively researching AI technologies and intends to take them all the way to practical deployment.
Scene-Adaptive Imaging Technology
Scene-Adaptive Imaging Technology
This was the one I personally found most fascinating. At the moment of capture, it divides the frame into 272 (16x17) regions and adjusts the camera mode locally to match the subject in each region. Since I use 360-degree cameras quite a bit myself, I know firsthand how hard it is to manage framing when a 360 camera indiscriminately captures everything all around you. For instance, if both dark and bright subjects appear in the frame at once, getting a clean exposure becomes nearly impossible. Tackling that problem through sheer brute force like this is super cool, and I can't wait until it reaches practical use.
And that wraps up my personal recap of the NHK STRL Open House. It's a great annual event where you can see steady incremental progress and get a sense of where things are headed.