Pointmap Association and Piecewise-Plane Constraint for Consistent and Compact 3D Gaussian Segmentation Field

3D Gaussian splatting (3DGS) offers significant advantages in real-time rendering, scene control, and editing within simultaneous localization and mapping (SLAM) systems. By using 3D Gaussian representations and splatting rendering techniques, 3DGS strikes a balance between speed and visual quality. However, many Gaussian ellipsoids and their associated parameters lead to high memory usage and storage demands. This poses challenges for large-scale applications and resource-constrained hardware such as lightweight GPUs. To address these issues, we introduce a novel, to our knowledge, large-scale compact 3D Gaussian splatting SLAM framework designed to minimize memory and storage requirements. First, we use dynamic masks and a geometric codebook to eliminate redundant Gaussians. Second, we propose a post-tuning and Gaussian pruning module that adjusts the Gaussian representations during post-tuning, enhancing scene reconstruction and further reducing the number of Gaussians. Experimental results on both RGB-D and monocular datasets show that our approach achieves a five-fold reduction in storage costs while delivering superior rendering quality compared to state-of-the-art algorithms., demonstrating improved tracking accuracy and reduced training time. Additionally, a preliminary evaluation under noisy outdoor-like conditions shows that the proposed framework retains competitive accuracy, indicating robustness beyond controlled indoor environments.

The dense simultaneous localization and mapping (SLAM) algorithm plays a crucial role in comprehending unknown environments. Recent studies have demonstrated the effectiveness of a dense SLAM system based on 3D Gaussian Splatting, enabling high-quality reconstruction and real-time rendering of scenes. However, many methods often rely on numerous Gaussian ellipsoids to represent the scene, leading to high memory consumption and time-consuming training process. To overcome this challenge, we propose a sampling method based on image gradients for initializing and expanding Gaussains that represent the scene map. This innovative method represents the scene map with compact 3D Gaussian distributions without degrading rendering quality. Moreover, to simplify keyframe management and improve the rendering performance in the mapping, we introduce a keyframe selection strategy that computes Gaussian covisibility according to the camera pose of the candidate frames in the mapping process. Our method achieves competitive performance compared to other advanced visual SLAM algorithms in camera pose estimation and map reconstruction, highlighting its efficiency in optimizing memory usage and improvement in algorithm performance.

In this paper, we propose to integrate superpixel-guided sampling into the framework of 3D Gaussian Splatting for novel view synthesis. Given a sequence of frames, where each frame is a pair of RGB-D image and the camera pose that captures the image, we first select the keyframes. Then, we decompose each keyframe image into superpixels, back-project each superpixel’s center into 3D space, and initialize a 3D Gaussian at the back-projected position. This superpixel-guided sampling produces a set of sparse but well-distributed Gaussians, which enables the optimization procedure to converge quickly. The experimental results show that with a significantly reduced computing cost, we can synthesize a novel view the quality of which is comparable to that generated by the state-of-the-art methods.

3D Gaussian Splatting (3DGS) has emerged as a mainstream for novel view synthesis, leveraging continuous aggregations of Gaussian functions to model scene geometry. However, 3DGS suffers from substantial memory requirements to store the large amount of Gaussians, hindering its efficiency and practicality. To address this challenge, we introduce GaussianSpa, an optimization-based simplification framework for compact and high-quality 3DGS. Specifically, we formulate the simplification objective as a constrained optimization problem associated with the 3DGS training. Correspondingly, we propose an efficient "optimizing-sparsifying" solution for the formulated problem, alternately solving two independent sub-problems and gradually imposing substantial sparsity onto the Gaussians in the 3DGS training process. We conduct quantitative and qualitative evaluations on various datasets, demonstrating the superiority of GaussianSpa over existing state-of-the-art approaches. Notably, GaussianSpa achieves an average PSNR improvement of 0.9 dB on the real-world Deep Blending dataset with 10× fewer Gaussians compared to the vanilla 3DGS. Our project page is available at https://noodle-lab.github.io/gaussianspa.

the scene representation, compared to 3DGS. Our work provides a comprehensive framework for 3D scene representation, achieving high performance, fast training, compactness, and real-time rendering. Our project page is available at https://maincold2.github.io/c3dgs/.

3D Gaussian Splatting holds significant potential for high-quality visual scene rendering. However, the large number of Gaussian primitives it requires poses challenges in memory consumption and practical deploy. Existing methods often rely on empirical criteria to prune Gaussians, which inevitably compromises visual quality. To address this, we propose Adversarial Pruning Networks (APNet), a framework that employs adversarial learning to balances the reduction of redundant Gaussians with the preservation of visual fidelity. APNet comprises a Gaussian Learning and Pruning Network (GLPN) and a Discriminative Network. GLPN incorporates the geometric information into the learning of Gaussians and prunes these Gaussians through a data-driven mask. Meanwhile, the Discriminative Network is trained to distinguish between synthesized and real images, acting as an adversary. Through adversarial pruning, APNet significantly reduces the number of Gaussians while rendering high-quality images. Extensive experiments on the Mip-NeRF360, Tanks & Temples, and Deep Blending datasets demonstrate that APNet achieves up to a 90% reduction in the original 3DGS while maintaining high rendering quality.

We calculate light scattering properties of random irregular particles of two different classes of shape, compact Gaussian random field particles and agglomerated debris particles, at size parameters X=50 and X=32. Surprisingly, very similar angular dependencies of all nonzero scattering matrix elements are obtained for both classes in the case of nonabsorbing material. For highly absorbing particles external scattering becomes dominant, which introduces a difference in the positive polarization due to different morphologies of their surfaces.

Relightable 4D avatar reconstruction which enables high fidelity and real-time rendering continues to be a crucial but challenging problem, especially from monocular videos. Previous NeRF-based 4D avatars enable photo-realistic relighting but are too slow for rendering, while point-based or mesh-based 4D avatars are efficient but have limited rendering quality. The recent success of 3D Gaussian Splatting, i.e., 3DGS, has inspired a series of impressive 4D Gaussian avatars, however, most of which only focus on faithful appearance reconstruction but are not relightable. To address such issues, this article proposes a new Relightable 4D Gaussian Avatar, i.e., RGAvatar, tailored for high fidelity relightable rendering from monocular videos. Our key idea is to introduce a new relightable 4D Gaussian representation, based on which we can directly perform high fidelity Physically Based Rendering, and an effective joint learning mechanism for compact 4D Gaussian reconstruction with SDF regulation and accurate materials and lighting decomposition. By comparing with previous state-of-the-art approaches, RGAvatar can significantly outperform previous approaches in relightable rendering quality and speed. To our best knowledge, RGAvatar contributes a new state-of-the-art 4D Gaussian avatar from monocular videos, which enables high fidelity relightable rendering in a quite efficient manner.

Accurate 3D instance segmentation is foundational for perception and decision making in embodied systems, yet prevailing approaches depend on densely annotated 3D point clouds, which are costly to acquire and difficult to scale across sensors and environments. We address this bottleneck with Fusion3DGS, an end-to-end, label efficient framework that couples 3D Gaussian Splatting with coordinated 2D-3D neural processing. From multi-view RGB images equipped only with 2D instance masks, our method optimizes a compact anisotropic Gaussian scene representation and performs instance reasoning via an occlusion aware cross-attention fusion stack. The weight-sharing lock imposes shape-consistent, gated coupling of early 2D and 3D kernels with conservative feedback and drift regularization to stabilize 2D-mask training and improve label efficiency, and a rendering consistency objective ties the Gaussian geometry to 2D supervision, enhancing boundary fidelity under occlusion and view changes. The ability to learn from widely available RGB data without dense 3D labels makes the approach practical for large-scale deployment.

Recent demonstrations [1] show that superlattice, mixed ferroelectric-antiferroelectric, fluorite-structure gate oxides with negative capacitance (NC) can go beyond state-of-the-art high-$\kappa$ HfO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> based equivalent oxide thickness (EOT) scaling [1–3]. However, the physical origin of NC in such gate stacks, especially a quantitative understanding of the gate capacitance, needs more detailed investigation. Here we present a comprehensive study of these gate stacks by combining pulsed-voltage and impedance measurements with compact modeling and 2D phase field simulations. We show that (i) a quantitative agreement between experiment and model can be reached and (ii) 2D effects play a critical role in determining what EOT can be attained. Our experimentally calibrated model provides insight into how substantially larger EOT scaling could be achieved.

This paper presents a dual-equalization-path (DEP) near field inductive coupling link for contactless vertical interconnect in high-performance 3D IC. The DEP receiver is demonstrated for the first time in near field inductive coupling vertical links to improve energy per bit and area efficiency. A feedback mechanism in the main path helps in signal recovery for improving the voltage margin. While a linear auxiliary path using passive pulse-reuse feedforward equalization helps in signal recovery for improving the timing margin. We also demonstrate how the receiver equalization architecture helps in enabling an energy-efficient and compact near field inductive coupling-based 3D IC. In experimental demonstration, the transceiver achieves 207 fJ/bit in 65nm CMOS process. The developed equalization architecture demonstrates a 34% improved FoM-3D performance compared to state-of-the-art contactless 3D links.

Radiance field based methods have recently revolutionized novel view synthesis of scenes captured with multi-view photos. A significant recent advance is 3D Gaussian Splatting (3DGS), which utilizes a set of 3D Gaussians to represent a radiance field, yielding high-fidelity results in real-time rendering. However, we have observed that 3DGS struggles to capture the necessary details in sparsely observed regions, where there is not enough gradient for effective split and clone operations. In this paper, we present a novel solution to address this limitation. Our key idea is to leverage segmentation information to identify poorly optimized regions within the 3D Gaussian representation. By applying split or clone operations on the corresponding 3D Gaussians in these regions, we aim to refine the spatial distribution of Gaussians and enhance the overall quality of high-fidelity 3D scene reconstruction. To further optimize the reconstruction process, we introduce two spatial regularization terms: repulsion loss and smoothness loss. These terms effectively minimize overlap and redundancy among Gaussians, reducing outliers in the synthesized geometry. By incorporating these regularization techniques, our approach achieves state-of-the-art performance in real-time novel view synthesis and significantly improves visibility in less observed regions, leading to a more compact and accurate 3D scene representation.

The recent development of feedforward 3D Gaussian Splatting (3DGS) presents a new paradigm to reconstruct 3D scenes. Using neural networks trained on large-scale multi-view datasets, it can directly infer 3DGS representations from sparse input views. Although the feedforward approach achieves high reconstruction speed, it still suffers from the substantial storage cost of 3D Gaussians. Existing 3DGS compression methods relying on scene-wise optimization are not applicable due to architectural incompatibilities. To overcome this limitation, we propose TinySplat, a complete feedforward approach for generating compact 3D scene representations. Built upon standard feedforward 3DGS methods, TinySplat integrates a training-free compression framework that systematically eliminates key sources of redundancy. Specifically, we introduce View-Projection Transformation (VPT) to reduce geometric redundancy by projecting geometric parameters into a more compact space. We further present Visibility-Aware Basis Reduction (VABR), which mitigates perceptual redundancy by aligning feature energy along dominant viewing directions via basis transformation. Lastly, spatial redundancy is addressed through an off-the-shelf video codec. Comprehensive experimental results on multiple benchmark datasets demonstrate that TinySplat achieves over 100× compression for 3D Gaussian data generated by feedforward methods. Compared to the state-of-the-art compression approach, we achieve comparable quality with only 6% of the storage size. Meanwhile, our compression framework requires only 25% of the encoding time and 1% of the decoding time.

3D Gaussian splatting has emerged as a promising technique for real-time scene representation, making interactive 3D segmentation increasingly important in scene manipulation. However, inconsistent results generated by 2D segmentation across different viewpoints pose significant challenges for learning 3D segmentation feature fields. The accuracy of 3D segmentation decreases substantially when cross-view 2D segmentation results conflict. To address this issue, we present Spatio-Temporal Feature-guided Learning for 3D Gaussian Segmentation (SFL-GS), an efficient interactive 3D segmentation framework. SFL-GS employs a novel Spatio-temporal Feature-guided Learning (SFL) strategy to capture spatio-temporally consistent features and masks from 2D segmentation results across different views, thereby coherently guiding the learning of 3D segmentation feature fields. To refine feature and mask consistency in complex scenes, particularly under severe occlusion, our framework develops an enhanced optimization strategy that integrates statistical filtering, dynamic scale growth, and edge-aware optimization. This approach results in clearer boundaries and significantly improves segmentation accuracy, even in challenging environments. Extensive experiments demonstrate that our method achieves superior accuracy in segmentation tasks, making it suitable for precise and efficient 3D segmentation requirements in real-world applications.

Automated quantification of plant-level development from multi-plant greenhouse scenes requires separating individual plants from shared scene-level reconstructions and quantifying organ-level development, a challenge that single-plant acquisition workflows do not directly address. This study presents an end-to-end phenotyping pipeline built on 3D Gaussian Splatting (3DGS) and a post-reconstruction extraction framework, LCR-GS, designed to isolate plant instances from full greenhouse scenes without scene-specific model retraining. LCR-GS integrates zero-shot 2D cues with multi-view lifting, geometric clustering, and chromatic refinement to convert large scene-level reconstructions (~2M Gaussians) into compact per-plant subsets (~16K Gaussians). Experiments on greenhouse-grown muskmelon at the early vegetative stage demonstrate high plant-extraction precision (0.933) and strong organ-level instance segmentation (mean AP50=0.924). Plant height and leaf count are validated against manual measurements (height R² = 0.98, RMSE=1.88 cm; leaf count R² = 0.86), whereas additional morphological traits, including leaf area, leaf area index, mean internode length, and stem node count, are reported as pipeline-derived descriptors for within-cohort comparison. By decoupling semantic inference from reconstruction, the pipeline reduces scene-scale data by over 99% and provides a practical route to derive compact per-plant 3D representations from multi-plant greenhouse imagery for downstream organ-level analysis.