Search

3D representation is essential to the significant advance of 3D generation with 2D diffusion priors. As a flexible representation, NeRF has been first adopted for 3D representation. With density-based volumetric rendering, it however suffers both intensive computational overhead and inaccurate mesh extraction. Using a signed distance field and Marching Tetrahedra, DMTet allows for precise mesh extraction and real-time rendering but is limited in handling large topological changes in meshes, leading to optimization challenges. Alternatively, 3D Gaussian Splatting (3DGS) is favored in both training and rendering efficiency while falling short in mesh extraction. In this work, we introduce a novel 3D representation, Tetrahedron Splatting (TeT-Splatting), that supports easy convergence during optimization, precise mesh extraction, and real-time rendering simultaneously. This is achieved by integrating surface-based volumetric rendering within a structured tetrahedral grid while preserving the desired ability of precise mesh extraction, and a tile-based differentiable tetrahedron rasterizer. Furthermore, we incorporate eikonal and normal consistency regularization terms for the signed distance field to improve generation quality and stability. Critically, our representation can be trained without mesh extraction, making the optimization process easier to converge. Our TeT-Splatting can be readily integrated in existing 3D generation pipelines, along with polygonal mesh for texture optimization. Extensive experiments show that our TeT-Splatting strikes a superior tradeoff among convergence speed, render efficiency, and mesh quality as compared to previous alternatives under varying 3D generation settings., Comment: Project page: https://fudan-zvg.github.io/tet-splatting/

Recent advancements in 3D generation are predominantly propelled by improvements in 3D-aware image diffusion models. These models are pretrained on Internet-scale image data and fine-tuned on massive 3D data, offering the capability of producing highly consistent multi-view images. However, due to the scarcity of synchronized multi-view video data, it remains challenging to adapt this paradigm to 4D generation directly. Despite that, the available video and 3D data are adequate for training video and multi-view diffusion models separately that can provide satisfactory dynamic and geometric priors respectively. To take advantage of both, this paper presents Diffusion$^2$, a novel framework for dynamic 3D content creation that reconciles the knowledge about geometric consistency and temporal smoothness from these models to directly sample dense multi-view multi-frame images which can be employed to optimize continuous 4D representation. Specifically, we design a simple yet effective denoising strategy via score composition of pretrained video and multi-view diffusion models based on the probability structure of the target image array. To alleviate the potential conflicts between two heterogeneous scores, we further introduce variance-reducing sampling via interpolated steps, facilitating smooth and stable generation. Owing to the high parallelism of the proposed image generation process and the efficiency of the modern 4D reconstruction pipeline, our framework can generate 4D content within few minutes. Notably, our method circumvents the reliance on expensive and hard-to-scale 4D data, thereby having the potential to benefit from the scaling of the foundation video and multi-view diffusion models. Extensive experiments demonstrate the efficacy of our proposed framework in generating highly seamless and consistent 4D assets under various types of conditions., Comment: Technical Report

Modeling dynamic, large-scale urban scenes is challenging due to their highly intricate geometric structures and unconstrained dynamics in both space and time. Prior methods often employ high-level architectural priors, separating static and dynamic elements, resulting in suboptimal capture of their synergistic interactions. To address this challenge, we present a unified representation model, called Periodic Vibration Gaussian (PVG). PVG builds upon the efficient 3D Gaussian splatting technique, originally designed for static scene representation, by introducing periodic vibration-based temporal dynamics. This innovation enables PVG to elegantly and uniformly represent the characteristics of various objects and elements in dynamic urban scenes. To enhance temporally coherent and large scene representation learning with sparse training data, we introduce a novel temporal smoothing mechanism and a position-aware adaptive control strategy respectively. Extensive experiments on Waymo Open Dataset and KITTI benchmarks demonstrate that PVG surpasses state-of-the-art alternatives in both reconstruction and novel view synthesis for both dynamic and static scenes. Notably, PVG achieves this without relying on manually labeled object bounding boxes or expensive optical flow estimation. Moreover, PVG exhibits 900-fold acceleration in rendering over the best alternative., Comment: Project page: https://fudan-zvg.github.io/PVG/

In this paper, we present a novel differentiable point-based rendering framework to achieve photo-realistic relighting. To make the reconstructed scene relightable, we enhance vanilla 3D Gaussians by associating extra properties, including normal vectors, BRDF parameters, and incident lighting from various directions. From a collection of multi-view images, the 3D scene is optimized through 3D Gaussian Splatting while BRDF and lighting are decomposed by physically based differentiable rendering. To produce plausible shadow effects in photo-realistic relighting, we introduce an innovative point-based ray tracing with the bounding volume hierarchies for efficient visibility pre-computation. Extensive experiments demonstrate our improved BRDF estimation, novel view synthesis and relighting results compared to state-of-the-art approaches. The proposed framework showcases the potential to revolutionize the mesh-based graphics pipeline with a point-based pipeline enabling editing, tracing, and relighting.

Neural Radiance Fields (NeRF) have been proposed for photorealistic novel view rendering. However, it requires many different views of one scene for training. Moreover, it has poor generalizations to new scenes and requires retraining or fine-tuning on each scene. In this paper, we develop a new NeRF model for novel view synthesis using only a single image as input. We propose to combine the (coarse) planar rendering and the (fine) volume rendering to achieve higher rendering quality and better generalizations. We also design a depth teacher net that predicts dense pseudo depth maps to supervise the joint rendering mechanism and boost the learning of consistent 3D geometry. We evaluate our method on three challenging datasets. It outperforms state-of-the-art single-view NeRFs by achieving 5$\sim$20\% improvements in PSNR and reducing 20$\sim$50\% of the errors in the depth rendering. It also shows excellent generalization abilities to unseen data without the need to fine-tune on each new scene.

We demonstrate an all-fiber spatiotemporal mode-locked laser based on graded-index multimode fiber. Due to the high damage threshold of graded-index multimode fiber devices, the output single-pulse energy reach up to 11.67 nJ. The beam quality at mode-locked state is significantly improved compared with continuous wave state as a result of the self-cleaning effect. By amplifying the pulse in the multimode fiber amplifier, high power pulses with high beam quality can be obtained in the all-fiber system based on graded-index MMFs., Comment: 8 pages, 5 figures

We firstly report mode-locked square-wave pulse in Yb-doped fiber laser based on graded index multimode fiber of reverse suturable absorption. By adjusting the pump power, the width of the square-wave can be tuned range from 350 ps to 52.6 ns with 3dB bandwidth spectrum of 0.73 nm. The supermode suppression ratio exceeds ~ 65 dB, which indicates excellent mode-locked operation state, and we also study the characteristic of pulse by the chirp measurement system. The mode-locked square-wave pulse fiber laser can sever as high power light source for industrial applications., Comment: 7 pages, 4 figures

Cylindrical vector beams (CVBs) have a wide range of applications owing to their particular polarization characteristics and optical field distributions. For the first time, an azimuthally polarized fiber laser without any polarization controller is proposed and demonstrated experimentally. The scheme is based on a self-designed ring-core fiber and transverse mode filter (TMF). The ring-core fiber can break the degeneracy of LP11 modes and make the TE01 mode propagate stably in the fiber laser. The TMF, which made from the ring-core fiber by depositing a layer of Aluminum on the cladding surface, can effectively suppress the modes other than TE01 mode. The fiber laser can stably operate at TE01 mode with a narrow 30dB linewidth of 0.18nm, which indicates the laser is polarization-maintained. This study opens a new avenue toward the true application of CVBs fiber lasers., Comment: 10 pages, 10 figures

We demonstrate an all-fiber mode-locked laser at all-normal dispersion with a saturable absorber, which is based on a section of grade-index multimode fiber with inner microcavity embedded in the splice points. The absorption modulation depth of this nonlinear saturable absorber is 2.2% in the suitable bent state. The fiber laser operates at central wavelength of 1035nm with average power of ~1.26mW, pulse-duration of 17.05 ps, repetition rate of 37.098MHz. This kind of saturable absorber has superiorities of low cost, high stability and high damage threshold, which may facilitate applications in optical sensing, material processing and light detection., Comment: 8 pages

We experimentally demonstrate a linear-cavity all-fiber passively Q-switching cylindrical vector beam (CVB) laser operating in high-order mode. This CVB fiber laser operates without any mode converter which always leads to high insertion loss, and it can realize high efficiency. In this fiber laser, the stable Q-switching pulse is achieved with a slope efficiency of 39%. By properly adjusting the polarization controllers, radially polarized and azimuthally polarized beams can be obtained. Our work proves the feasibility of achieving the stable Q-switching CVB pulse with high-order mode directly oscillating, and it may have an enormous potential for enhancing the efficiency., Comment: 4 pages

A novel mode-locked fiber laser based on offset splicing technology between two kinds of graded index multimode fibers is proposed and experimentally demonstrated. The offset splice spot structure acts as a saturable absorber via nonlinear multimodal interference mechanism. The absorption modulation depth of the device is 2.8%, with a saturation intensity of 2.5 MW/cm2. The fiber laser operates at a stable mode-locked state when the pump power reaches 114 mW. The laser oscillates at 1039 nm with a pulsewidth of 17.6 ps, a repetition rate of 32.619 MHz, and a high signal noise ratio of 65 dB. The fiber laser is superior in terms of low cost, high damage threshold, and easy fabrication, thus providing many potential applications in ultrafast photonics and industrial processing., Comment: 8 pages.8 figures

Roe scheme is known for its good performance in moderate-Mach-number flows. However, this scheme and its extended versions suffers from many disastrous problems, such as non-physical behavior, global cut-off, and checkerboard problems, for incompressible flows; and shock instability, expansion shock, and positively non-conservative problems for hypersonic flows. In this paper, non-physical behavior problem, checkerboard problem, and main reason of shock instability problem are due to that the Roe scheme cannot identify multi-dimensional incompressible and compressible flows when normal Mach number on the cell face tends to zero, and then leads to incorrect cross modifications. Positively non-conservative problem is also identified as another important reason for shock instability. Therefore, Mach number and an assistant pressure-density-varying detector are introduced into the Roe scheme to judge compressibility, positivity condition is satisfied by a simple modification with minimal numerical dissipation increases and even with possible decreases in numerical dissipation, the mechanism of the preconditioned Roe scheme is introduced to suppress checkerboard problem, and modified entropy fix and the rotated Riemann solver is combined with complementary advantages as an assistant improvement for better robust. Based on above improvements and previous developments for global cut-off and expansion shock problems, an improvement Roe scheme for all Mach-number flow (Roe-AM) is proposed to simultaneously overcome nearly all well-known drawbacks of the classical Roe scheme. The Roe-AM scheme is simple, easy to implement, computationally low-cost, robust, good extensibility, and free of empirical parameters essentially, with increasing minimal numerical dissipation.

A common defect of the Roe scheme is the production of non-physical expansion shock and shock instability. An improved method with several advantages was presented to suppress the shock instability. However, this method cannot prevent expansion shock and is incompatible with the traditional curing method for expansion shock. Therefore, the traditional curing mechanism is analyzed. The discussion explains the effectiveness of the traditional curing method and identifies several defects, one of which leads to incompatibility between curing the shock instability and expansion shock. Consequently, a new improved Roe scheme is proposed in this study. This scheme is concise, easy to implement, low computational cost, and robust. More importantly, the scheme can simultaneously cure the shock instability and expansion shock without additional costs.