Implicit neural rendering, especially Neural Radiance Field (NeRF), has shown great potential in novel view synthesis of a scene. However, current NeRF-based methods cannot enable users to perform user-controlled shape deformation the While existing works have proposed some approaches modify radiance field according user's constraints, modification is limited color editing or object translation and rotation. In this paper, we propose method that allows controllable on implicit representation...

In this paper, we present an implicit surface reconstruction method with 3D Gaussian Splatting (3DGS), namely 3DGSR, that allows for accurate intricate details while inheriting the high efficiency and rendering quality of 3DGS. The key insight is to incorporate signed distance field (SDF) within Gaussians modeling, enable alignment joint optimization both SDF Gaussians. To achieve this, design coupling strategies align associate Gaussians, allowing unified enforcing constraints on With...

We propose a new method for realistic human motion transfer using generative adversarial network (GAN), which generates video of target character imitating actions source character, while maintaining high authenticity the generated results. tackle problem by decoupling and recombining posture information appearance both characters. The innovation our approach lies in use projection reconstructed 3D model as condition GAN to better maintain structural integrity results different poses....

Neural Radiance Fields (NeRFs) have shown great potential for tasks like novel view synthesis of static 3D scenes. Since NeRFs are trained on a large number input images, it is not trivial to change their content afterwards. Previous methods modify provide some control but they do support direct shape deformation which common geometry representations triangle meshes. In this paper, we present NeRF editing method that first extracts mesh representation the inside NeRF. This can be modified by...

Novel view synthesis for dynamic scenes is still a challenging problem in computer vision and graphics. Recently, Gaussian splatting has emerged as robust technique to represent static enable high-quality real-time novel synthesis. Building upon this technique, we propose new representation that explicitly decomposes the motion appearance of into sparse control points dense Gaussians, respectively. Our key idea use points, significantly fewer number than learn compact 6 DoF transformation...

Pose transfer of human videos aims to generate a high-fidelity video target person imitating actions source person. A few studies have made great progress either through image translation with deep latent features or neural rendering explicit 3D features. However, both them rely on large amounts training data realistic results, and the performance degrades more accessible Internet due insufficient frames. In this paper, we demonstrate that dynamic details can be preserved even when trained...

We propose a new method for realistic human motion transfer using generative adversarial network (GAN), which generates video of target character imitating actions source character, while maintaining high authenticity the generated results. tackle problem by decoupling and recombining posture information appearance both characters. The innovation our approach lies in use projection reconstructed 3D model as condition GAN to better maintain structural integrity results different poses....

In this paper, we present an implicit surface reconstruction method with 3D Gaussian Splatting (3DGS), namely 3DGSR, that allows for accurate intricate details while inheriting the high efficiency and rendering quality of 3DGS. The key insight is incorporating signed distance field (SDF) within Gaussians to enable them be aligned jointly optimized. First, introduce a differentiable SDF-to-opacity transformation function converts SDF values into corresponding Gaussians' opacities. This...

Video representation is a long-standing problem that crucial for various down-stream tasks, such as tracking,depth prediction,segmentation,view synthesis,and editing. However, current methods either struggle to model complex motions due the absence of 3D structure or rely on implicit representations are ill-suited manipulation tasks. To address these challenges, we introduce novel explicit representation-video Gaussian -- embeds video into Gaussians. Our proposed models appearance in...

Recently, Gaussian splatting has emerged as a robust technique for representing 3D scenes, enabling real-time rasterization and high-fidelity rendering. However, Gaussians' inherent radial symmetry smoothness constraints limit their ability to represent complex shapes, often requiring thousands of primitives approximate detailed geometry. We introduce Deformable Radial Kernel (DRK), which extends into more general flexible framework. Through learnable bases with adjustable angles scales, DRK...

Implicit neural rendering, especially Neural Radiance Field (NeRF), has shown great potential in novel view synthesis of a scene. However, current NeRF-based methods cannot enable users to perform user-controlled shape deformation the While existing works have proposed some approaches modify radiance field according user's constraints, modification is limited color editing or object translation and rotation. In this paper, we propose method that allows controllable on implicit representation...