S$^2$GS: Structured Sparse Gaussian Streaming for Efficient Free-Viewpoint Video Reconstruction on Edge-IoT Devices

Yiwei Li; Jiannong Cao; Weixun Gao; Rui Cao; Songye Zhu; Yinfeng Cao; Mingjin Zhang

cs.CVAug 20, 2026

S $^2$ GS: Structured Sparse Gaussian Streaming for Efficient Free-Viewpoint Video Reconstruction on Edge-IoT Devices

Authors: Yiwei Li, Jiannong Cao, Weixun Gao, Rui Cao, Songye Zhu, Yinfeng Cao, Mingjin Zhang

Organizations: Department of Computing, The Hong Kong Polytechnic University, Hong Kong SAR, China · Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University, Hong Kong SAR, China

Abstract

Streaming reconstruction of Free-Viewpoint Videos (FVVs) supports immersive Internet of Things (IoT) services, such as telepresence and digital twin visualization. Existing methods suffer from high per-frame optimization time and large storage footprints, limiting deployment on resource-constrained Edge-IoT devices. To address these challenges, we propose Structured Sparse Gaussian Streaming (S $^2$ GS), an FVV reconstruction framework that exploits structure-aware temporal sparsity to selectively update Gaussian residuals, enabling efficient streaming without compromising visual fidelity. In the spatial domain, a streaming octree hierarchically organizes Gaussian residuals, capturing spatial correlations that guide residual updates. In the temporal domain, a structured gating mechanism, comprising hierarchical feature propagation (HFP) and Gumbel-Sigmoid sampling, converts hierarchical dynamic cues into sparse residual update decisions under differentiable optimization. A multi-level discrete scheme is further adopted to provide fine-grained control over residual updates while preserving intricate dynamic details. Extensive experiments across consumer GPUs, industrial edge IoT devices, and a physical telepresence testbed demonstrate that S $^2$ GS consistently reduces per-frame optimization time and storage footprint while maintaining competitive visual quality. Compared with QUEEN, S $^2$ GS reduces per-frame optimization time by 59% and storage costs by 85% on an RTX 4090 GPU. On the Jetson AGX Orin, S $^2$ GS delivers the highest rendering throughput (60+ FPS) and the lowest energy consumption among the evaluated methods, demonstrating its potential for deployment in resource-constrained systems.

Figures & tables

Fig. 1: Schematic comparison of Gaussian residual update paradigms. Top: Existing GS-based streaming methods perform primitive-wise dense residual updates. Bottom: S 2 GS exploits spatially correlated temporal sparsity to select a small number of hierarchical regions and perform region-wise sparse residual updates only within active regions.

Fig. 2: Overview and performance of the proposed structured sparse Gaussian streaming framework (S²GS). S²GS delivers high-quality FVV reconstruction with improved efficiency in per-frame training, rendering, and storage. (a) Quantitative comparison on the N3DV dataset, where PSNR, per-frame training time, storage cost (circle size), and rendering throughput (color) are jointly compared. (b) Visualization of sparse residual updates on the challenging Flame Salmon scene. S 2 GS-fast concentrates updates on dynamic regions, using only 4.52% active Gaussians and 4.37% active gates for efficient FVV streaming. (c) Edge deployment on the N3DV dataset at 1/8 resolution, where S²GS-edge outperforms baseline methods in visual quality, storage, per-frame training time, and rendering throughput.

Fig. 3: Overview of the proposed S²GS framework. (a) The streaming octree representation is initialized using root Gaussian primitives from the first frame and subsequently represents FVVs with multi-resolution grids, enabling hierarchical allocation of Gaussian residuals and efficient point queries for FVV streaming. (b) A structured gating mechanism is introduced to induce sparse Gaussian residual updates through hierarchical feature propagation, differentiable sampling with Gumbel-Sigmoid, and multi-level (ML) discretization using a straight-through estimator (STE). (c) A sparse regularization loss is incorporated to enable efficient end-to-end optimization of structured gates and sparse residual updates.

Fig. 4: Analysis of different STE variants. (a) Continuous soft gates and their discretized counterparts produced by Binary STE and the proposed Multi-Level (ML) STE. Unlike Binary STE, ML-STE preserves finer-grained gate states after discretization. (b) Number of active gates over training epochs for a representative FVV frame. Binary STE rapidly leads to gate collapse, whereas ML-STE maintains stable sparse residual updates throughout training. (c) Visual comparison on the same FVV frame as in (b), where ML-STE achieves better reconstruction quality than Binary STE, particularly in the dynamic region.

Methods

PSNR (dB)

\uparrow

/ SSIM

\uparrow

#Gaussians (k)

\downarrow

Storage (MB)

\downarrow

Train (Sec)

\downarrow

Render (FPS)

\uparrow

Coffee Martini

Cook Spinach

Cut Beef

Flame Salmon

Flame Steak

Sear Steak

Average

Offline

K-Planes † [ 40 ]

29.99 / 0.925

32.60 / 0.942

31.82 / 0.943

30.44 / 0.924

32.38 / 0.957

32.52 / 0.955

31.63 / 0.941

1.04

21.6

0.3

NeRFPlayer † [ 11 ]

31.53 / 0.951

30.56 / 0.929

29.35 / 0.908

31.65 / 0.940

31.93 / 0.950

29.13 / 0.908

30.69 / 0.931

17.1

72.0

0.05

4DGaussians [ 26 ]

30.10 / 0.935

31.15 / 0.958

32.60 / 0.951

30.21 / 0.934

33.49 / 0.961

32.64 / 0.961

31.70 / 0.950

192

0.29

7.92

Grid4D [ 28 ]

28.69 / 0.919

32.90 / 0.957

33.61 / 0.958

29.86 / 0.930

32.98 / 0.963

33.49 / 0.965

31.92 / 0.949

202

0.16

16.4

127

STG [ 23 ]

28.61 / 0.916

33.18 / 0.952

33.52 / 0.954

29.48 / 0.918

33.64 / 0.960

33.89 / 0.961

32.05 / 0.944

434

0.67

15.1

149

TABLE I: Quantitative comparisons on the N3DV dataset at 1/2 resolution. For all methods, per-frame metrics are averaged across all frames. Results for methods marked with † are taken from the original papers. All experiments are conducted on an NVIDIA RTX 4090 GPU. Red and blue denote the best and second-best results, respectively, within each category.

Fig. 5: Qualitative comparison with SOTA methods [ 15 , 19 , 16 , 12 ] on diverse datasets [ 10 , 9 , 42 ] . Our method captures fine-grained details presented in indoor and outdoor scenes, particularly for objects with intricate motions such as poured liquid, hands, knives, and stuffed toys.

	Methods	Meet Room Dataset							ENeRF Outdoor Dataset
	Methods	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#Gaussians (k) $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#Gaussians (k) $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
Online	3DGStream [ 15 ]	29.30	0.942	0.188	185	4.10	4.77	260	25.91	0.856	0.129	2182	8.77	12.96	117
	4DGC [ 19 ]	30.41	0.947	0.180	208	0.77	24.52	155	24.52	0.845	0.135	1275	0.93	37.71	102
	HiCoM [ 16 ]	29.57	0.944	0.182	152	0.39	3.91	377	22.12	0.749	0.280	761	1.74	4.43	385
	QUEEN-s [ 12 ]	31.95	0.960	0.159	133	0.23	2.92	414	26.21	0.839	0.177	282	0.95	2.94	411
	QUEEN-l [ 12 ]	32.21	0.962	0.152	135	0.29	3.51	397	26.13	0.843	0.154	354	1.32	3.82	358

TABLE II: Quantitative comparison on Meet Room and ENeRF Outdoor datasets. For all methods, metrics are averaged across all frames. Results for methods marked with † are taken from the original papers. All experiments are conducted on an NVIDIA RTX 4090 GPU. Red and Blue indicate the best and second best, respectively, within each category.

Fig. 6: Trend comparison of PSNR, storage cost, training time, and rendering throughput for S²GS-full and QUEEN-l on the Vrheadset scene from the Meet Room dataset. The triangle ( $\blacktriangle$ ) denotes the metric value at the first frame.

Incremental Design	PSNR (dB) $\uparrow$	#Gaussians (k) $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
Baseline	29.61	412	410k	3.09	13.7	199
+ STE	31.89	352	107k	2.37	12.9	221
+ Sparse Reg.	32.25	331	16k	1.10	12.7	234
+ Stream. Octree	32.71	101	2.4k	0.29	4.24	485
+ HFP (S²GS-full)	32.76	101	679	0.11	4.12	482

TABLE III: Ablation study on the N3DV dataset. Red and Blue indicate the best and second best, respectively, within each category.

Structure		Sparsity		MeetRoom Dataset						ENeRF Outdoor Dataset
Stream. Octree	HFP	STE	Sparse Reg.	PSNR (dB) $\uparrow$	#Gaussians (k) $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$	PSNR (dB) $\uparrow$	#Gaussians (k) $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
-	-	-	-	31.69	129	129k	0.67	2.86	383	25.12	439	398k	4.17	4.03	285
-	-	$\checkmark$	$\checkmark$	31.86	127	13k	0.22	2.73	387	25.69	382	51k	1.41	3.66	299
$\checkmark$	$\checkmark$	-	-	32.33	64	17k	0.38	2.18	496	25.80	138	39k	1.77	2.45	532
$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	32.46	64	401	0.08	1.96	513	25.94	138	1.9k	0.20	2.28	541

TABLE IV: 2

\times

2 factorial ablation study of structure and sparsity. Structure denotes the proposed streaming octree with HFP, and sparsity denotes STE with sparse regularization. Red and Blue indicate the best and second-best results, respectively.

Configuration	PSNR (dB) $\uparrow$	LPIPS $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
HFP (S²GS-full)	25.94	0.124	1.9k	0.20	2.28	541
FT, $\tau=0.50$	25.82	0.128	1.8k	0.19	2.35	537
FT, $\tau=0.80$	25.72	0.133	1.6k	0.18	2.39	526
FT, $\tau=0.90$	25.33	0.141	1.2k	0.15	2.37	539
FT, $\tau=0.95$	25.12	0.150	707	0.13	2.32	544

TABLE V: Propagation Strategy Analysis on the ENeRF Outdoor Dataset. Red and Blue indicate the best and second best, respectively. FT denotes fixed-threshold pruning.

Configuration	PSNR (dB) $\uparrow$	SSIM $\uparrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
STE: Binary	29.85	0.934	2	0.06	4.05	478
STE: ML (S²GS-full)	32.76	0.952	679	0.11	4.12	482

TABLE VI: Comparison of discrete gating strategies on the N3DV dataset. Red indicates the best result within each category. ML denotes the multi-level scheme.

Fig. 7: Effect of hierarchical feature propagation (HFP) on the Flame Salmon scene. (a) With HFP, the number of sparse active gates is reduced by nearly one order of magnitude at a lower ratio. (b) HFP effectively identifies dynamic regions while suppressing noise.

Resolution	Methods	Quality & Efficiency							Resource Consumption
Resolution	Methods	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#Gaussians (k) $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$	CPU (%) $\downarrow$	GPU MEM (GB) $\downarrow$	RAM (GB) $\downarrow$	Power (W) $\downarrow$	Energy (J) $\downarrow$
1/4	3DGStream [ 15 ]	33.21	0.953	0.095	386	8.21	63.28	21.22	38.23	1.82	0.78	14.56	921.36
	HiCoM [ 16 ]	31.25	0.950	0.085	194	0.49	45.61	24.81	208.0	1.53	1.34	13.53	617.10
	QUEEN-s [ 12 ]	32.40	0.955	0.085	260	0.60	32.27	32.35	25.86	1.91	1.29	14.29	461.14
	S²GS-edge (Ours)	33.07	0.958	0.083	82	0.14	18.61	49.73	41.98	1.42	1.32	11.21	208.62
1/8	3DGStream [ 15 ]	32.48	0.952	0.063	397	8.17	38.90	28.77	40.64	1.71	0.74	15.55	604.90

TABLE VII: Quantitative comparison on the Jetson AGX Orin with the multi-resolution N3DV dataset. Energy per frame is computed as average power multiplied by per-frame training time. Red and Blue indicate the best and second best, respectively.

Fig. 8: Qualitative comparison of different propagation strategies. FT-based variants progressively blur high-motion areas as $\tau$ increases, whereas HFP preserves fine details and sharpness. Results indicate that HFP effectively retains informative dynamic responses, particularly in regions with subtle but meaningful local motion.

Fig. 9: Physical testbed with IoT application of the proposed method.

Fig. 10: Qualitative comparison with SOTA methods [ 15 , 16 , 12 ] on the physical testbed. S 2 GS-edge achieves clearer visual details, demonstrating its practical feasibility for edge-assisted IoT applications.

Appendix figures & tables17 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	Scene	Root Resolution	Log Base	$\lambda_{1}$	$\lambda_{2}$
N3DV	Coffee Martini	$2^{12}$	2.0	0.01	0.01
N3DV	Cook Spinach	$2^{9}$	2.0	0.01	0.01
N3DV	Cut Roasted Beef	$2^{9}$	2.0	0.01	0.01
N3DV	Flame Salmon	$2^{12}$	2.0	0.01	0.01
N3DV	Flame Steak	$2^{9}$	2.0	0.01	0.01
N3DV	Sear Steak	$2^{9}$	2.0	0.01	0.01

Appendix

TABLE VIII: Per-scene hyperparameter configurations. The physical testbed and Campus dataset are additional evaluations reported in the appendix.

Streaming Octree		N3DV Dataset					MeetRoom Dataset
HRA	ESQ	PSNR (dB) $\uparrow$	#Gaussians (k) $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	PSNR (dB) $\uparrow$	#Gaussians (k) $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$
-	-	32.25	331	16k	1.10	12.7	31.89	127	13k	0.22	2.83
$\checkmark$	-	32.74	122	2.7k	0.34	4.58	32.40	71	2.4k	0.16	2.26
$\checkmark$	$\checkmark$	32.71	101	2.4k	0.29	4.24	32.44	64	2.2k	0.14	2.02

Appendix

TABLE IX: Ablation study of the streaming octree in terms of reconstruction quality, parameter efficiency, and training speed. Red indicates the best result. HRA and ESQ denote hierarchical residual allocation and efficient streaming query, respectively. Experiments are conducted without the HFP module.

Resolution	PSNR (dB) $\uparrow$	#Gaussians (k) $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
$2^{7}$	32.39	35	154	0.05	1.97	526
$2^{8}$	32.46	64	401	0.08	1.96	513
$2^{9}$	32.59	89	828	0.11	2.02	505
$2^{10}$	32.41	102	999	0.12	2.04	498
$2^{11}$	32.36	106	974	0.12	2.00	501

Appendix

TABLE X: Effect of octree root resolution on the MeetRoom dataset.

Log Base	PSNR (dB) $\uparrow$	#Gaussians (k) $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
$1.2$	32.23	122	1.2k	0.16	2.18	495
$1.5$	32.46	64	401	0.08	1.96	513
$2.0$	32.29	48	218	0.06	2.02	501
$2.5$	32.33	36	165	0.05	1.95	519
$3.0$	32.20	28	117	0.04	1.91	525

Appendix

TABLE XI: Effect of anchor selection on the MeetRoom dataset.

	N3DV Dataset					ENeRF Outdoor Dataset
Configuration	PSNR (dB) $\uparrow$	SSIM $\uparrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	PSNR (dB) $\uparrow$	SSIM $\uparrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$
S 2 GS-full (Ours)	32.76	0.952	679	0.11	4.12	25.94	0.863	1.9k	0.20	2.28
w/o Prog. Train.	32.68	0.951	755	0.14	4.15	26.02	0.861	2.1k	0.24	2.33

Appendix

TABLE XII: Ablation study on the effect of progressive training. Red indicates the best result.

Fig. 11: Visualizations of rendered images from S 2 GS-full and the variant without progressive training. Progressive training facilitates the recovery of dynamic objects with fine-grained motions.

	First-Frame Metric		Streaming Performance
Epochs	PSNR (dB) $\uparrow$	Train (Sec) $\downarrow$	PSNR (dB) $\uparrow$	LPIPS $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
50	26.92	9.15	29.72	0.148	0.16	3.64	475
100	28.18	19.75	29.80	0.139	0.16	3.65	460
150	28.93	27.68	29.83	0.135	0.16	3.70	462
200	29.26	37.99	29.95	0.134	0.17	3.74	477
250	28.96	46.69	30.07	0.133	0.17	3.82	469

Appendix

TABLE XIII: Effect of the first-frame training budget on the Coffee Martini scene. Epochs indicate the number of optimization epochs used for the first frame.

	First-Frame Metric		Streaming Performance
Noise	PSNR (dB) $\uparrow$	Train (Sec) $\downarrow$	PSNR (dB) $\uparrow$	LPIPS $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
$\alpha=0$	33.54	29.91	33.96	0.134	0.07	4.20	498
$\alpha=10$	33.30	30.22	33.74	0.135	0.07	4.23	492
$\alpha=20$	30.89	29.84	33.55	0.138	0.07	4.21	489
$\alpha=30$	28.93	28.86	33.06	0.144	0.07	4.26	495

Appendix

TABLE XIV: Effect of Gaussian noise degradation during first-frame training on the Cook Spinach scene.

Fig. 12: Controlled illumination changes on the Cut Roasted Beef scene with a time-varying global gain $g_{t}\in[0.5,1.5]$ .

Confidence Estimator	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$
Uniform Random	32.33	0.964	0.147	801	0.12	2.09
Constant-One	32.20	0.963	0.143	12k	0.63	2.48
Constant-Zero	25.13	0.919	0.201	0	0.04	1.91
Min-Max VGD	26.03	0.923	0.197	1	0.04	1.92
Robust VGD (Ours)	32.46	0.964	0.147	401	0.08	1.96

Appendix

TABLE XV: Ablation study of different dynamic confidence estimators on the MeetRoom dataset. Min-Max VGD denotes raw VGD normalized using min-max scaling.

Illumination Gain	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$
$g_{t}=1.0$	34.12	0.958	0.133	744	0.07	4.10
$g_{t}\in[0.8,1.2]$	34.28	0.958	0.132	713	0.07	4.11
$g_{t}\in[0.6,1.4]$	34.41	0.958	0.128	753	0.08	4.16
$g_{t}\in[0.5,1.5]$	34.98	0.958	0.123	777	0.08	4.21

Appendix

TABLE XVI: Robustness analysis under controlled illumination changes on the Cut Roasted Beef scene.

Injected Noise	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$
$\alpha=0$	30.26	0.939	0.109	235	0.18	4.17
$\alpha=0.1$	30.18	0.939	0.109	181	0.16	4.19
$\alpha=0.2$	30.25	0.939	0.110	179	0.16	4.19
$\alpha=0.3$	30.19	0.939	0.108	182	0.16	4.18
$\alpha=0.5$	30.13	0.938	0.109	189	0.16	4.20

Appendix

TABLE XVII: Robustness analysis under injected gradient noise on the Flame Salmon scene.

$\lambda_{1}$	$\lambda_{2}$	PSNR (dB) $\uparrow$	LPIPS $\downarrow$	#Active Gates $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$
Effect of $\lambda_{1}$ for Gate Decision
0	0.01	25.60	0.122	12k	0.79	2.44	533
0.01	0.01	25.94	0.124	1.9k	0.20	2.28	541
0.02	0.01	25.84	0.127	1.2k	0.15	2.23	539
0.03	0.01	25.78	0.128	952	0.14	2.23	547
Effect of $\lambda_{2}$ for Offset Magnitude

Appendix

TABLE XVIII: Sensitivity analysis of the regularization weights on the ENeRF Outdoor dataset. The default setting is

\lambda_{1}=0.01

and

\lambda_{2}=0.01

Scenes	Methods	Quality & Efficiency							Resource Consumption
Scenes	Methods	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#Gaussians (k) $\downarrow$	Storage (MB) $\downarrow$	Train (Sec) $\downarrow$	Render (FPS) $\uparrow$	CPU (%) $\downarrow$	GPU MEM (GB) $\downarrow$	RAM (GB) $\downarrow$	Power (W) $\downarrow$	Energy (J) $\downarrow$
Install RAM	3DGStream [ 15 ]	19.38	0.700	0.301	695	8.47	52.84	16.24	31.46	2.60	0.72	15.91	840.96
	HiCoM [ 16 ]	17.50	0.659	0.289	293	0.70	22.54	36.06	175.7	1.41	1.09	13.62	306.88
	QUEEN-s [ 12 ]	18.61	0.758	0.183	290	0.89	7.83	48.13	45.55	1.49	1.15	15.81	123.77
	S²GS-edge (Ours)	19.04	0.792	0.133	88	0.26	4.60	59.37	65.63	1.09	1.20	12.91	59.38
Pick Blocks	3DGStream [ 15 ]	17.85	0.697	0.324	518	8.43	50.52	21.63	35.43	2.08	0.75	15.20	767.90

Appendix

TABLE XIX: Quantitative comparison on the NVIDIA Jetson AGX Orin in the case study. Energy per frame is computed as average power multiplied by per-frame training time. Red and Blue indicate the best and second best, respectively.

Fig. 13: Frames from our captured multi-view video ( Pick Blocks scene). We use 8 camera views for training and the center view for evaluation. The acquisition setup reflects several practical constraints commonly encountered in industrial FVV streaming, including limited viewpoints, low input resolution, and cross-view illumination variations.

Fig. 14: Qualitative comparison with SOTA methods [ 16 , 15 , 12 ] on the Unplug Device scene. S2GS-edge achieves clearer results, demonstrating its practical feasibility for edge-assisted IoT applications.

Fig. 15: Qualitative results on the Campus dataset [ 46 ] . S 2 GS may produce blurred or incomplete reconstructions under occlusion/disocclusion, large object motion with out-of-support displacement, and fine-scale stochastic dynamics, as observed in pedestrians emerging from pillars, a fast-moving football with large displacement, and transient splashing fountain droplets.

Streaming 3D Reconstruction

Streaming 3D reconstruction incrementally builds geometrically consistent scene maps from incoming images or video under tight memory and latency limits.

See all 99 papers

Struct-GStream: Towards Efficient Free-Viewpoint Video Streaming at Low-Bitrates with Structured 3D Gaussians
Aug 2, 2026Han Jiao, Jiakai Sun, Lei Zhao +43DGS Compression 3D Gaussian Splatting
College of Computer Science and Technology, Zhejiang University, China · Chinese Academy of Sciences, Beijing, China · University of Science and Technology Beijing, China
ResStreamGS: Residual-Sensitive Dual-Stream Gaussian Splatting for Mobile Volumetric Video Streaming
Oct 15, 2026Haoran Liu, Di Wu, Wang Luo +33D Gaussian Splatting 3DGS Compression
DecoGS: Adaptive Static-Dynamic Decoupling of 3D Gaussians for Free-Viewpoint Video Streaming
Sep 15, 2026Idil Sulo, Alexey Supikov, Ilke Demir +13D Gaussian Splatting Streaming 3D Reconstruction
University of Bonn, Almetra · V3DEO · Cauth AI +1

Aug 2, 2026J/K move · Enter open · S save

Struct-GStream: Towards Efficient Free-Viewpoint Video Streaming at Low-Bitrates with Structured 3D Gaussians

Han Jiao, Jiakai Sun, Lei Zhao, Wei Xing+3

College of Computer Science and Technology, Zhejiang University, China · Chinese Academy of Sciences, Beijing, China · University of Science and Technology Beijing, China

3DGS Compression 3D Gaussian Splatting Streaming 3D Reconstruction Dynamic Scene Reconstruction

Open paper AskPDF arXiv

Constructing photorealistic Free-Viewpoint Videos (FVVs) of dynamic scenes from a set of posed 2D images has been an intriguing yet challenging task in computer vision. Methods based on neural rendering achieve high-fidelity image quality in FVV construction. However, most of these methods are unable to achieve real-time rendering and often require complete video sequences to train. Despite the existence of some online training methods capable of rendering FVVs in real time, they struggle to meet the requirements for storage and training time for downstream applications. To overcome this problem, we propose Struct-GStream, which can achieve efficient FVV streaming using structured 3D Gaussians (3DGs). Specifically, we introduce dynamic anchor points to generate structured 3DGs to construct basic scenes and model approximate scene movements based on the assumption of local rigidity in object motion. Besides, we introduce a global free 3DGs patching strategy involving free 3DGs' generation, pruning, and optimization to patch and model deficient areas and emerging objects. Our method achieves fast training at low bitrates while maintaining high rendering quality. Extensive experiments demonstrate that Struct-GStream significantly outperforms existing online training methods for FVV construction in terms of training time, storage, and rendering quality while maintaining competitive rendering speed.

Abstract

Streaming 3D Reconstruction

Explore similar work

Struct-GStream: Towards Efficient Free-Viewpoint Video Streaming at Low-Bitrates with Structured 3D Gaussians

ResStreamGS: Residual-Sensitive Dual-Stream Gaussian Splatting for Mobile Volumetric Video Streaming

DecoGS: Adaptive Static-Dynamic Decoupling of 3D Gaussians for Free-Viewpoint Video Streaming