AEGIS: Preserving privacy of 3D Facial Avatars with Adversarial Perturbations

MetroGS has been introduced as a novel Gaussian Splatting framework aimed at achieving efficient and stable reconstruction of geometrically accurate high-fidelity large-scale scenes, particularly in complex urban environments. This method leverages a distributed 2D Gaussian Splatting representation and incorporates a structured dense enhancement scheme to address challenges in sparse regions and improve reconstruction completeness.

A new method called NVGS has been proposed to enhance 3D Gaussian Splatting by learning viewpoint-dependent visibility functions for occlusion culling, addressing the limitations posed by the semi-transparent nature of Gaussians. This approach utilizes a shared MLP across instances and integrates neural queries into an instanced software rasterizer, improving rendering efficiency and image quality.

The Observer Actor (ObAct) framework has been introduced, enhancing active vision imitation learning by allowing a robotic observer to optimize visual observations for an actor arm. This system utilizes wrist-mounted cameras to create a 3D Gaussian Splatting representation, enabling the observer to find optimal camera poses and improve the execution of policies by the actor arm.

A new approach called Group Training has been introduced to enhance the efficiency of 3D Gaussian Splatting (3DGS), a technique known for its high-fidelity scene reconstruction. This method organizes Gaussian primitives into groups, significantly improving training speed and rendering quality, achieving up to 30% faster convergence in various scenarios.

A new framework called Physics-Informed Deformable Gaussian Splatting (PIDG) has been proposed to enhance 3D Gaussian Splatting (3DGS) by incorporating physics-driven motion patterns. This method treats Gaussian particles as Lagrangian material points with time-varying parameters, improving the accuracy of dynamic scene representation through supervision by 2D optical flow and the Cauchy momentum residual.

A recent study introduces Frequency-Adaptive Sharpness Regularization (FASR) to enhance the generalization capabilities of 3D Gaussian Splatting (3DGS) in few-shot scenarios, addressing its tendency to overfit to sparse observations. This approach reframes the training objective to improve convergence towards better generalization solutions, particularly for novel view synthesis.

A new paper titled 'DensifyBeforehand: LiDAR-assisted Content-aware Densification for Efficient and Quality 3D Gaussian Splatting' introduces a method that enhances 3D Gaussian Splatting (3DGS) by combining sparse LiDAR data with monocular depth estimation from RGB images. This approach aims to improve the initialization of 3D scenes and reduce artifacts associated with adaptive density control.

A new framework for sparse-input novel view synthesis has been introduced, focusing on test-time natural video completion. This approach utilizes pretrained video diffusion models to generate plausible in-between views from limited scene glimpses, enhancing spatial coherence and scene reconstruction through an iterative feedback loop.