TED-4DGS: Temporally Activated and Embedding-based Deformation for 4DGS Compression
Cheng-Yuan Ho, He-Bi Yang, Jui-Chiu Chiang, Yu-Lun Liu, Wen-Hsiao Peng
2025-12-11
Summary
This paper introduces a new method, called TED-4DGS, for efficiently representing and compressing videos as 3D scenes. It builds upon recent advances in representing static 3D scenes with 'Gaussian Splatting' and extends it to handle changes over time.
What's the problem?
Existing methods for representing dynamic 3D scenes either use too many details that quickly disappear, making them inefficient, or they deform the 3D scene in ways that aren't naturally tied to how time progresses. Essentially, they struggle to find a good balance between accurately showing movement and keeping the file size manageable. Compressing these dynamic 3D scenes effectively, meaning getting a good visual quality with a small file size, hasn't been fully explored.
What's the solution?
TED-4DGS solves this by using a system of 'anchors' – think of them as key points in the 3D scene. Each anchor has settings that control when it appears and disappears in the video. It also uses a 'deformation bank' which is a collection of ways to subtly change the shape of each anchor over time, and each anchor picks the deformation that works best for it. To make the file size even smaller, the method uses a smart way to predict the properties of these anchors and how they relate to each other, using something called an 'implicit neural representation' and 'autoregressive modeling'.
Why it matters?
This work is important because it's one of the first to focus on compressing dynamic 3D scenes in a way that balances visual quality and file size. This could lead to more efficient ways to store and share complex 3D videos, potentially impacting fields like virtual reality, augmented reality, and even filmmaking.
Abstract
Building on the success of 3D Gaussian Splatting (3DGS) in static 3D scene representation, its extension to dynamic scenes, commonly referred to as 4DGS or dynamic 3DGS, has attracted increasing attention. However, designing more compact and efficient deformation schemes together with rate-distortion-optimized compression strategies for dynamic 3DGS representations remains an underexplored area. Prior methods either rely on space-time 4DGS with overspecified, short-lived Gaussian primitives or on canonical 3DGS with deformation that lacks explicit temporal control. To address this, we present TED-4DGS, a temporally activated and embedding-based deformation scheme for rate-distortion-optimized 4DGS compression that unifies the strengths of both families. TED-4DGS is built on a sparse anchor-based 3DGS representation. Each canonical anchor is assigned learnable temporal-activation parameters to specify its appearance and disappearance transitions over time, while a lightweight per-anchor temporal embedding queries a shared deformation bank to produce anchor-specific deformation. For rate-distortion compression, we incorporate an implicit neural representation (INR)-based hyperprior to model anchor attribute distributions, along with a channel-wise autoregressive model to capture intra-anchor correlations. With these novel elements, our scheme achieves state-of-the-art rate-distortion performance on several real-world datasets. To the best of our knowledge, this work represents one of the first attempts to pursue a rate-distortion-optimized compression framework for dynamic 3DGS representations.