New Compression Trick Shrinks 3D Video Files by Up to 72%

Imagine a video where you can move around inside the scene instead of just watching a flat rectangle. That's what "3D Gaussian splatting" does: it rebuilds a real place — a stadium, a living room, a concert — out of millions of tiny fuzzy blobs, so you can look at it from any angle. The problem is size. A few seconds of moving 3D footage can be far larger than a normal video file, which makes streaming it to a headset slow, expensive, and data-hungry.

A team of researchers took a fresh angle on shrinking these files. Their insight is simple to picture: instead of storing each split-second separately, they stack a group of frames into one big bundle and label each blob with the frame it belongs to. That converts the repetition across time into repetition across space — and ordinary, already-existing compression tools can squash that. Importantly, the method doesn't need to know how the 3D scene was built.

The numbers are striking. Across six standard test clips, their approach cut file sizes by 42% to 72% compared with compressing each frame on its own — for footage where the software successfully tracks objects moving through the scene. When tracking fails, the gains mostly vanish, ranging from a 3.5% improvement to a slight 5% loss. One popular codec, HGSC, didn't play nicely with the technique at all. It's also a research paper, not a product you can download and use today, though the code is public.

So what does this mean for you? Bigger, smoother 3D experiences on the same internet connection. Think watching a basketball game from any seat, touring a house you're considering buying, or attending a concert in a headset — without the stuttering and buffering that plague today's 3D video. It also means companies pay less to store and send that content, which usually trickles down to cheaper services and lower data bills.