: Transactions on Machine Learning Research (TMLR) 2025 Key Contributions
: The method first reconstructs a video into a 3DGS representation and then integrates it into zero-shot video editors to maintain stability across frames. VIDE-3GS.mp4
: It uses separate sets of Gaussians ( Frg-3DGS and Bkg-3DGS ) to represent moving objects and static backgrounds independently. : Transactions on Machine Learning Research (TMLR) 2025