ECCV 2026

MoBa-GS: Learning a Spatially-Varying Motion Basis
over a Dynamic Canonical Space for 4D Reconstruction

1Monash University
~163 FPS
Real-time rendering
(time-invariant caching)
~18 min
Training time
(order-of-magnitude speedup)
~11 MB
Storage footprint
(compact representation)
SfM-Free
Random initialization
no COLMAP required
SOTA
Fidelity
Surpasses prior art on
D-NeRF & NeRF-DS
Teaser Video or Image → static/videos/teaser.mp4  or  static/images/teaser.jpg

MoBa-GS faithfully captures complex non-rigid dynamics by introducing a structural inversion: a low-frequency dynamic canonical space that models coarse scene motion, coupled with a Spatially-Varying Motion Basis that decomposes local kinematics per Gaussian. Time-invariant caching yields real-time rendering at >163 FPS with only ~11 MB storage and ~18 min training — all without SfM initialization.

Abstract

Faithfully capturing the intricate relationship between motion and geometry in dynamic scenes is essential for 4D reconstruction. Recent state-of-the-art methods rely on monolithic deformation networks or global time-basis factorizations that struggle to represent complex, non-rigid topological changes. We propose MoBa-GS, a framework that resolves this entanglement by introducing a structural inversion: a spatially-factorized motion field coupled with adaptive geometric optimization. First, our model learns a low-frequency dynamic canonical space to represent coarse scene motion. Next, it decomposes complex, non-rigid motion into a Spatially-Varying Motion Basis of local kinematics, predicted from the canonical geometry, which is then linearly combined using dynamic blending weights. This formulation directly acts as an implicit neural scaffold, recovering both geometry and motion from random initialization, thereby removing the dependency on SfM point cloud priors. This design is further augmented with motion-guided densification and positional annealing to reduce geometry overfitting. Extensive experiments show that our framework surpasses prior state-of-the-art methods in reconstruction fidelity. Enabled by time-invariant caching, MoBa-GS requires an order-of-magnitude shorter training time (~18 minutes), a compact storage (~11 MB), and achieves real-time rendering speeds (~163 FPS). Our work establishes a new foundation for high-fidelity, efficient 4D representations without relying on explicit geometric priors.

Method

Pipeline Overview Figure → static/images/framework.jpg

Figure 1: MoBa-GS Overview. (Stage 1) A low-frequency dynamic canonical space is learned to capture coarse, global scene motion. (Stage 2) A Spatially-Varying Motion Basis decomposes the residual non-rigid motion into local kinematic subspaces, one per Gaussian, predicted from the canonical geometry. Dynamic blending weights linearly combine the basis vectors into a full per-Gaussian deformation field. (Training) Motion-guided densification and positional annealing progressively refine geometry, removing the need for SfM initialization. At test time, time-invariant feature caching decouples rendering from network evaluation, achieving >163 FPS.

Qualitative Comparisons

Novel view synthesis comparisons on D-NeRF and NeRF-DS datasets.

D-NeRF — Flame Salmon
MoBa-GS (Ours)
static/videos/ours_flame_salmon.mp4
SC-GS
static/videos/scgs_flame_salmon.mp4
Deformable-GS
static/videos/defgs_flame_salmon.mp4
4D-GS
static/videos/4dgs_flame_salmon.mp4
D-NeRF — Jumping Jacks
MoBa-GS (Ours)
static/videos/ours_jumping_jacks.mp4
SC-GS
static/videos/scgs_jumping_jacks.mp4
Deformable-GS
static/videos/defgs_jumping_jacks.mp4
4D-GS
static/videos/4dgs_jumping_jacks.mp4
D-NeRF — Hell Warrior
MoBa-GS (Ours)
static/videos/ours_hell_warrior.mp4
SC-GS
static/videos/scgs_hell_warrior.mp4
Deformable-GS
static/videos/defgs_hell_warrior.mp4
4D-GS
static/videos/4dgs_hell_warrior.mp4
NeRF-DS — Real-world Scenes
MoBa-GS (Ours)
static/videos/ours_nerfds.mp4
SC-GS
static/videos/scgs_nerfds.mp4
DynMF
static/videos/dynmf_nerfds.mp4
4D-GS
static/videos/4dgs_nerfds.mp4

Quantitative Results

Table 1 — D-NeRF Synthetic Dataset

Quantitative table (D-NeRF) → static/images/table_dnerf.png

Comparison on D-NeRF synthetic dataset (PSNR / SSIM / LPIPS). MoBa-GS surpasses prior art in reconstruction fidelity while delivering an order-of-magnitude reduction in both training time and storage.

Table 2 — NeRF-DS Real-World Dataset

Quantitative table (NeRF-DS) → static/images/table_nerfds.png

Mean comparison across all NeRF-DS scenes. MoBa-GS handles real-world reflective and non-rigid surfaces without SfM priors.

Visual Comparisons

D-NeRF qualitative comparison → static/images/qual_dnerf.png

Figure 2: D-NeRF qualitative results. MoBa-GS recovers sharper boundaries and reduces floaters in fast-moving regions compared to SC-GS, Deformable-GS, and 4D-GS.

NeRF-DS qualitative comparison → static/images/qual_nerfds.png

Figure 3: NeRF-DS qualitative results. The Spatially-Varying Motion Basis captures per-region motion heterogeneity on real-world reflective and deformable surfaces with high fidelity.

Ablation Studies

Ablation table / figure → static/images/ablation.png

Figure 4: Ablation studies. We ablate the dynamic canonical space, the Spatially-Varying Motion Basis, motion-guided densification, and positional annealing. Each component provides measurable, complementary improvements in both fidelity and efficiency.

BibTeX

@inproceedings{tan2026mobags,
  author    = {Tan, Guan Yuan and Pal, Arghya and Rajanala, Sailaja
               and Phan, Rapha{\"{e}}l C-W and Ting, Chee-Ming},
  title     = {MoBa-GS: Learning a Spatially-Varying Motion Basis
               over a Dynamic Canonical Space for 4D Reconstruction},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026},
}