DimFlow improves dynamic scene reconstruction with faster, accurate rendering
DimFlow: Bidirectional Grid-Expert Learning for Dynamic Gaussian Reconstruction
Computer Vision and Pattern Recognition
Summary
Reconstructing changing 3D scenes requires capturing details about how shapes and appearances change over time. The authors propose DimFlow, a method that combines a grid of features with a specialized model to better keep dynamic information consistent and accurate. This hybrid learning approach helps generate clearer, faster images of the moving scenes. Compared to previous methods, DimFlow achieves better image quality without slowing down rendering speed.
What this means in practice
- •For game developers: Enable real-time rendering of complex dynamic 3D scenes with improved quality and speed using DimFlow’s coordinated feature learning.$Commercial implications: Provides a practical method to enhance dynamic scene graphics in games, potentially integrated into rendering engines for better visuals at scale.
- •For augmented reality developers: Support realistic, fast updates of dynamic environments in AR applications by applying DimFlow’s dynamic Gaussian reconstruction approach.
Authors
Gabriel Adineera, Jinoh Kim
Abstract
Dynamic Gaussian reconstruction requires compact scene features to support coordinated changes in geometry and appearance. A factorized grid provides efficient local queries, but accurate interpolation alone does not ensure that the resulting dynamic attributes remain compatible across space and time. We introduce DimFlow, a framework that learns these features through bidirectional interaction between a multi-resolution grid and a coordinate-based expert branch. Feature-level adversarial alignment connects their distributions, adaptive fusion transfers complementary information, and activation-masked residual feedback refines the grid during training. Four attribute experts organize deformation, rotation, opacity, and temporal information before Gaussian rendering. In the aggregate comparison across D-NeRF, HyperNeRF, and DAVIS, DimFlow achieves 39.73 dB PSNR, improving on 4D-GS by 1.92 dB while matching its reported 85 FPS rendering speed. The model occupies 18.9 MB, compared with 18.3 MB for 4D-GS. Component and attribute-expert ablations reveal complementary contributions from feature alignment, adaptive exchange, and temporal processing.