E2EGS: Event-to-Edge Gaussian Splatting for Pose-Free 3D Reconstruction

Yunsoo Kim; Changki Sung; Dasol Hong; Hyun Myung

cs.CVMar 16, 2026

E2EGS: Event-to-Edge Gaussian Splatting for Pose-Free 3D Reconstruction

Authors: Yunsoo Kim, Changki Sung, Dasol Hong, Hyun Myung

Organizations: Urban Robotics Lab, School of Electrical Engineering, KAIST

Abstract

The emergence of neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS) has advanced novel view synthesis (NVS). These methods, however, require high-quality RGB inputs and accurate corresponding poses, limiting robustness under real-world conditions such as fast camera motion or adverse lighting. Event cameras, which capture brightness changes at each pixel with high temporal resolution and wide dynamic range, enable precise sensing of dynamic scenes and offer a promising solution. However, existing event-based NVS methods either assume known poses or rely on depth estimation models that are bounded by their initial observations, failing to generalize as the camera traverses previously unseen regions. We present E2EGS, a pose-free framework operating solely on event streams. Our key insight is that edge information provides rich structural cues essential for accurate trajectory estimation and high-quality NVS. To extract edges from noisy event streams, we exploit the distinct spatio-temporal characteristics of edges and non-edge regions. The event camera's movement induces consistent events along edges, while non-edge regions produce sparse noise. We leverage this through a patch-based temporal coherence analysis that measures local variance to extract edges while robustly suppressing noise. The extracted edges guide structure-aware Gaussian initialization and enable edge-weighted losses throughout initialization, tracking, and bundle adjustment. Extensive experiments on both synthetic and real datasets demonstrate that E2EGS achieves superior reconstruction quality and trajectory accuracy, establishing a fully pose-free paradigm for event-based 3D reconstruction.

Figures & tables

Figure 1: Edge-guided reconstruction framework. Our pipeline extracts robust edges from consecutive event maps (Sec. 3.2 ), initializes edge-aware Gaussians (Sec. 3.3 ), and applies edge-guided losses during joint optimization of 3D scene representation and camera trajectory (Sec. 3.4 ). Depth sampling generates edge-guided Gaussians along viewing rays using inverse depth distribution, while surface sampling randomly initializes complementary Gaussians to cover texture-less regions. $k$ -NN stands for $k$ -nearest neighbors.

Method	room0			room2			office0			office2			office3
Method	PSNR↑	SSIM↑	LPIPS↓	PSNR↑	SSIM↑	LPIPS↓	PSNR↑	SSIM↑	LPIPS↓	PSNR↑	SSIM↑	LPIPS↓	PSNR↑	SSIM↑	LPIPS↓
EvGGS [ 29 ]	17.57	0.32	0.68	11.28	0.31	0.69	14.34	0.22	0.66	14.82	0.32	0.73	15.51	0.28	0.68
Event-3DGS ∗ [ 8 ]	22.27	0.84	0.30	16.39	0.68	0.36	15.97	0.64	0.51	18.19	0.60	0.48	17.82	0.84	0.31
IncEventGS [ 11 ]	23.54	0.86	0.20	23.25	0.76	0.27	26.53	0.49	0.44	21.27	0.76	0.29	19.21	0.82	0.19
IncEventGS †	19.81	0.83	0.24	22.58	0.73	0.29	27.72	0.51	0.40	24.48	0.83	0.26	20.04	0.85	0.18
Ours	23.86	0.87	0.19	23.01	0.77	0.26	28.01	0.52	0.41	24.86	0.83	0.25	20.75	0.85	0.19

Table 1: NVS performance on Replica dataset. Our method achieves superior reconstruction quality solely using event data. † denotes no depth supervision and ∗ denotes that the method uses camera poses obtained through E2VID [ 24 ] for frame reconstruction followed by COLMAP [ 26 ] for pose estimation.

Method	Synthetic (Replica)					Real-world (TUM-VIE)
Method	room0	room2	office0	office2	office3	1d	3d	6dof	desk	desk2
DEVO [ 16 ]	0.271	0.381	0.287	0.134	0.356	0.23	1.00	1.82	0.55	0.75
ESVO2 [ 20 ]	-	-	-	-	-	1.08	1.51	1.49	0.89	2.65
IncEventGS [ 11 ]	0.051	0.071	0.085	0.041	0.058	2.19	1.62	0.70	0.15	0.41
IncEventGS †	6.817	0.446	0.698	0.513	1.379	2.58	4.48	8.24	0.78	96.59
Ours	0.049	0.065	0.078	0.070	0.077	1.12	0.65	0.58	0.12	0.40

Table 2: Absolute trajectory error (ATE). RMSE (cm) across synthetic Replica and real TUM-VIE sequences. Our method achieves competitive performance solely using event data.

Figure 2: ATE with respect to the length of the sequence.

Method	Edge loss	Edge init	Depth init	ATE (cm)
IncEventGS [ 11 ]	✗	✗	✓	0.37
IncEventGS †	✗	✗	✗	6.62
w/ Edge loss	✓	✗	✗	0.50
w/ Edge init	✗	✓	✗	0.29
Ours	✓	✓	✗	0.28

Table 3: Ablation study on edge components.

Panel of Figure 3 — Figure 3: Qualitative results on Replica dataset. Red boxes highlight regions of interest for comparison. Our method produces sharper boundaries and cleaner surfaces compared with baselines. IncEventGS shows failures including wave-like artifacts, missing details, and indistinct boundaries. IncEventGS † exhibits severe reconstruction failures due to accumulated trajectory estimation errors.

Panel of Figure 4 — Figure 4: Impact of trajectory error on reconstruction quality. (a) Ground truth. (b) IncEventGS exhibits multiple failure modes: spatial misalignment causing viewpoint shifts and blurred regions in distant areas beyond initial coverage (top), and objects disappearing (bottom). (c) IncEventGS † fails to reconstruct scenes due to severe trajectory errors. (d) Our method achieves accurate reconstruction with sharp textures and correct spatial alignment.

Panel of Figure 5 — Figure 5: Effect of edge-guided loss. Baseline (top) vs. ours with edge loss but random initialization (bottom) at early and final training stages. Each pair shows rendered image (left) and depth map (right). Depth maps show that our edge-guided loss enables faster convergence.

Panel of Figure 6 — Figure 6: Effect of edge ratio on reconstruction quality. Red boxes highlight comparison regions. (a) Ground truth. (b) Without edge guidance, fine details such as dot patterns on the back wall are lost. (c-d) Moderate edge ratios achieve optimal reconstruction with well-preserved details. (e) Excessive edge emphasis causes reconstruction degradation due to insufficient surface coverage.

Edge Ratio ( $r_{\text{edge}}$ )	0.0	0.1	0.3	0.5	0.7	1.0
ATE (cm)	5.68	0.41	0.40	1.78	5.53	11.93

Table 4: Ablation study on edge ratio (

r_{\text{edge}}

3D Reconstruction

3D reconstruction recovers a scene’s shape, appearance, and camera motion from images, video, depth, or other sensor measurements.

See all 950 papers

Aug 9, 2026

EvTrajGS: Accurate and Efficient 3D Gaussian Splatting from Unposed Event Streams

Event cameras, with high temporal resolution, high dynamic range, and asynchronous sensing characteristics, have shown great potential for dense 3D reconstruction. Traditional reconstruction methods based on off-the-shelf pose estimates achieve high efficiency but produce low-fidelity results, as inaccurate pose initialization introduces cumulative reconstruction errors. In contrast, recent SLAM-style methods stabilize joint pose-scene optimization through incremental tracking and mapping, yielding higher reconstruction fidelity at the expense of considerable computational overhead. To address this trade-off, this paper presents EvTrajGS, an accurate and efficient 3D Gaussian Splatting framework for unposed event streams. Our method enables reliable joint pose-scene optimization initialized from coarse pose priors, eliminating the need for computationally expensive SLAM-style pipelines. EvTrajGS parameterizes camera motion as a continuous-time trajectory initialized from discrete camera poses, providing a unified representation for pose refinement. We then aggregate adjacent trajectory states into a temporally coupled pose, promoting temporally consistent pose updates during joint optimization. Additionally, we introduce a loss-reweighted event sampling strategy to adaptively emphasize temporally under-reconstructed intervals. Extensive experiments on both synthetic and real-world datasets demonstrate that EvTrajGS outperforms state-of-the-art methods in terms of both geometric reconstruction quality and pose estimation accuracy, achieving 3.8 dB higher PSNR, 0.1 higher SSIM, and over 40% lower ATE RMSE while retaining high computational efficiency.

Zixuan Chen, Jiakai Zhang, Junhao Dong +4

School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou 510006, Guangdong, China; and with the Guangdong Province Key Laboratory of Information Security Technology, Guangzhou 510006, Guangdong, China; and also with the Key Laboratory of Machine Intelligence and Advanced Computing, Ministry of Education, Guangzhou 510006, Guangdong, China · College of Computing and Data Science, Nanyang Technological University, Singapore, Singapore · School of Computing and Information Technology, Great Bay University, Dongguan 523000, Guangdong, China

3D Gaussian Splatting 3D Reconstruction Camera Pose Estimation

Ask PDF

Aug 9, 2026

ERF-GS: Reconstructing Fast Motion from Disjoint Event-RGB Viewpoints

Deep learning-driven representations such as neural radiance fields (NeRFs) and 3D Gaussian splatting (3DGS) have revolutionized the field of dynamic 3D scene reconstruction with improved visual precision and scalability. However, the reconstruction of fast-moving objects remains a challenge; existing methods based on conventional frame-based videos often struggle in scenarios such as sports events and animal videography. We propose an event-RGB fusion Gaussian splatting (ERF-GS) framework that integrates event information into both optimization and densification stages of the Gaussian splatting pipeline, taking advantage of novel event sensors with high frame-rate. Unlike many other event-assisted scene reconstruction methods, ERF-GS was developed using realistic simulation settings and realizes event-based learning detached from RGB inputs. This design enables its application beyond straightforward synthetic data into the realm of natural video with complex layout, low frame rates and severe motion blur. Our experiments show that ERF-GS outperforms both the 4DGS baseline and the concurrent E-D3DGS on different variants of the Neu3D and Nvidia datasets which include blurry RGB frames and disjoint RGB-event viewpoints. Our code is available at https://github.com/andrewbxy/ERF-GS.

Xiaoyang Bai, Zhenyang Li, Weiwei Xu +2

Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong SAR, China · State Key Lab of CAD&CG, Zhejiang University, Hangzhou, China

3D Gaussian Splatting 3D Reconstruction Dynamic Scene Reconstruction Multimodal Sensor Fusion

Ask PDF

Jul 17, 2026

Event3R: Asynchronous-to-Global 3D Reconstruction from Event Camera via Spatial-Temporal Feature Aggregation

Robust 3D reconstruction is essential for robotics and embodied perception. Recent feed-forward approaches such as DUSt3R have demonstrated impressive progress in dense 3D reconstruction from RGB images, achieving global geometric consistency and strong generalization. However, extending such dense 3D reconstruction to event cameras remains challenging due to their asynchronous, sparse, and highly dynamic nature, as well as the lack of large-scale, well-labeled datasets. In this work, we introduce Event3R, a feed-forward framework that directly maps asynchronous event streams to globally consistent 3D point clouds. Event3R represents incoming events as spatial-temporal voxels, enabling time-aware feature integration through a temporal attention module that enhances the module's temporal feature learning. To further strengthen temporal representation learning and reduce reliance on labeled data, we propose a Masked Bin Modeling (MBM) strategy for self-supervised pre-training, enabling robust temporal representation learning with minimal labeled data, and retain it as an auxiliary fine-tuning objective. In addition, contrastive alignment and consistency regularization losses are incorporated during fine-tuning to reinforce structural correspondence and temporal coherence across views. Extensive experiments on both synthetic and real-world benchmarks demonstrate that Event3R achieves robust, temporally consistent, and globally aligned 3D reconstructions, significantly outperforming existing event-based methods.

Jian Huang, Haotian Shen, Xinhao Lou +3

Zhejiang University, Hangzhou, Zhejiang, China · Westlake University, Hangzhou, Zhejiang, China

Event-Based Vision 3D Reconstruction Feed-Forward 3D Reconstruction

Ask PDF

Abstract

3D Reconstruction

Explore similar work

EvTrajGS: Accurate and Efficient 3D Gaussian Splatting from Unposed Event Streams

ERF-GS: Reconstructing Fast Motion from Disjoint Event-RGB Viewpoints

Event3R: Asynchronous-to-Global 3D Reconstruction from Event Camera via Spatial-Temporal Feature Aggregation