Learning from Automatic 3D Gaussian Splatting Edits for Online Bayesian Scene Change Detection

Changrui Zhu; Simon Julier

cs.CVOct 20, 2026

Learning from Automatic 3D Gaussian Splatting Edits for Online Bayesian Scene Change Detection

Authors: Changrui Zhu, Simon Julier

Organizations: University College London

Abstract

Online scene change detection requires integrating incoming observations to determine which regions have changed relative to a reference scene. We update change beliefs as new images arrive using generalised Bayesian evidence fusion. However, the prior probability of change and evidence models are unknown beforehand, and numerical Bayesian inference can be costly. We address these challenges by learning offline from automatic edits to a reference 3D Gaussian Splatting (3DGS) model. These edits generate labelled training views without manual annotations. Together with unchanged reference observations, they provide supervision for learning scene-specific evidence models and estimating the effective pixel prior and response statistics. Online prediction combines current evidence with historical support, while closed-form local Bayes factors, parameterized by the learned response statistics, update an accumulated 3D Gaussian field. Object-guided refinement produces the final change masks. Our method achieves state-of-the-art accuracy on PASLCD and CCS3D, with $67.64\%$ and $75.82\%$ mIoU, respectively. On PASLCD, the complete pipeline runs at $17.2$ ,FPS over complete query sequences, the fastest among the evaluated online methods. Project page: https://changruizhu96.github.io/work/lego-scd/

Figures & tables

Figure 1: Online evidence fusion pipeline. Current evidence $C_{t}$ combines mapped image evidence $P_{t}$ and Gaussian-aggregated support $S_{t}$ . Fusion with prior $\pi$ and history $H_{t}$ from $F_{t-1}$ gives $Q_{t}^{\mathrm{pixel}}$ ; refinement produces $Q_{t}^{\mathrm{full}}$ . Thresholding at $0.5$ gives masks $\widehat{Y}_{t}^{\mathrm{pixel}}$ and $\widehat{Y}_{t}^{\mathrm{full}}$ , respectively. Local Bayes factors $B_{tg}$ update $F_{t-1}$ to $F_{t}$ for subsequent queries. $\oplus$ adds log odds.

Figure 2: Automatic reference editing and change evidence learning. Top : Reference 3DGS edits (recolouring, removal, translation) and relighting synthesize labelled change pairs without manual annotation. Bottom : Five scene-specific components are fitted in the order required by their inputs (grey arrows): response suppressor $\mathcal{T}$ , LLR mapping $\ell_{\mathrm{img}}$ , unchanged-response model $(\mu_{g},\tau_{g}^{2},c)$ , pixel prior $\pi$ , and contour filter $\ell_{\mathrm{contour}}$ .

Figure 3: Visual results on representative scenes. Columns show reference renderings, queries, ground truth, O-SCD(R) (offline), and our complete method (online). correct detections; false positives; missed changes. Numbers give per-frame foreground IoU from one complete sequence run. Additional visual results, including remaining errors, are shown in Appendix H .

Table 4

Pixel pipeline			Output refinement
Configuration	mIoU $\uparrow$	F1 $\uparrow$	Configuration	mIoU $\uparrow$	F1 $\uparrow$
Full pixel pipeline	60.36	73.80	Full method	67.64	78.87
w/o response suppression	58.62	72.17	w/o contour filtering	63.15	75.78
w/o likelihood-ratio mapping	59.23	73.12	w/o object completion	62.56	75.32
w/o Gaussian-aggregated support	58.47	72.27	w/o contour image weights	67.38	78.67
w/o historical support	57.81	71.67

Table 3: Independent component ablations on 20 PASLCD instances (%). Each component is removed separately from the corresponding complete pipeline with the decision threshold fixed at

0.5

. Removing the pixel prior is equivalent to using a neutral prior,

\pi=0.5

Method	Mode	Post-frontend (ms) $\downarrow$	End-to-end (ms) $\downarrow$	FPS $\uparrow$
O-SCD(R) †	offline	169.41	201.34	4.97
O-SCD(O)	online	37.65	69.33	14.42
Ours (pixel)	online	4.79	35.81	27.93
Ours (full)	online	25.64	58.23	17.17

Table 4: Runtime on PASLCD (RTX 5090; 20 instances). Post-frontend covers detection, state updates, and refinement. Timing definitions and protocols are given in Appendix G.5 .

Appendix figures & tables15 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 4: Selected examples from the editing subset, showing recolouring, removal, and translation in four scenes. Each row uses the same crop across the real reference image, synthetic query, relit synthetic query, and edit mask. White marks regions labelled changed; translation includes both original and displaced locations. Relighting preserves these labels. Condition-specific ignore masks are omitted.

Scene	Inst.	O-SCD(O) CVPR 2026		O-SCD(R) CVPR 2026		Ours pixel		Ours full
		mIoU	F1	mIoU	F1	mIoU	F1	mIoU	F1
Cantina	1	50.24	65.86	62.05	76.05	57.84	72.68	62.79	76.23
	2	49.38	65.16	54.75	70.19	55.29	70.48	58.98	73.16
Garden	1	49.58	65.52	52.28	67.07	68.38	80.43	75.31	85.33
	2	51.54	67.78	56.62	72.27	67.44	80.37	74.77	85.51
Lounge	1	54.74	70.38	60.04	74.91	68.96	81.41	75.18	85.63

Appendix

Table 5: PASLCD comparison using our evaluator over 20 scene-instances (%, higher is better). O-SCD(O) is online; O-SCD(R) includes offline refinement. Ours (pixel) is the pixel output; Ours (full) adds object-guided mask refinement. Our results average two complete runs; Total summarizes all 20 instances. Cell backgrounds mark first, second, and third place for each row and metric; these values are bold, and ties at the displayed precision share a rank.

	Original pairing		Cross-instance
Output	mIoU	F1	mIoU	F1
Ours (pixel)	60.36	73.80	59.93	73.39
Ours (full)	67.64	78.87	67.15	78.41

Appendix

Table 6: Same-scene cross-instance reference reuse on PASLCD (%). Original pairing uses each instance’s own reference; cross-instance pairing uses the other instance’s reference and fitted parameters. Both average two runs of 20 sequences.

		Pixel output		Full output
Scene	Direction	mIoU	F1	mIoU	F1
Cantina	$1\to 2$	57.82	72.61	62.87	76.17
Cantina	$2\to 1$	54.89	70.08	57.88	72.29
Garden	$1\to 2$	67.21	79.65	73.76	84.35
Garden	$2\to 1$	66.99	80.03	74.53	85.33
Lounge	$1\to 2$	68.94	81.48	74.88	85.47

Appendix

Table 7: Cross-instance results for all 20 directed PASLCD pairs (%). Direction denotes reference instance

\to

query instance. Pixel and full outputs use the same fixed reference reconstruction and fitted parameters.

Figure 5: Fitted pixel priors on PASLCD. Each scene has a distinct colour; circles and squares denote its two instances, joined by a dashed line. The vertical dotted line marks the median over all 20 instances. The horizontal axis uses a logarithmic scale.

Change area	Frames	Instances	mIoU $\uparrow$	FPR $\downarrow$	FNR $\downarrow$	mIoU $\uparrow$	FPR $\downarrow$	FNR $\downarrow$
			Pixel output			Refined output
$<1\%$	88	12	49.50	0.64	36.13	54.83	0.58	37.74
$1$ – $5\%$	298	18	61.18	0.96	18.18	70.11	0.60	16.69
$5$ – $10\%$	84	12	66.39	0.97	24.88	71.12	0.67	22.48
$10$ – $20\%$	29	4	67.35	1.67	25.27	71.10	1.44	22.38
$20$ – $30\%$	1	1	63.44	5.19	23.50	66.87	4.82	20.33

Appendix

Table 8: PASLCD performance by ground-truth change area. Metrics average two runs over the same 500 queries (%). The

20

–

30\%

group contains only one frame; no evaluated frame has

\geq 30\%

change area.

	Pixel output		Full output
Strength	mIoU $\uparrow$	F1 $\uparrow$	mIoU $\uparrow$	F1 $\uparrow$
$K=0$	60.04	73.54	66.98	78.45
$K=1$ (default)	60.36	73.80	67.64	78.87
$K=2$	60.40	73.83	67.57	78.85

Appendix

Table 9: Relighting strength on PASLCD, averaged over 20 instances (%). All settings retain the same training records;

K

scales the lighting perturbation about the identity transform.

Historical support	mIoU $\uparrow$	F1 $\uparrow$
None	57.81	71.67
Temporal averaging	55.94	70.27
Ours	60.36	73.80

Appendix

Table 10: Historical support on PASLCD, averaged over 20 instances (%, higher is better). Ours uses Bayes-factor accumulation. All results use pixel outputs without refinement.

Method	Frontend	Detection	Refinement	Input/loop	Total
O-SCD(O)	31.62	37.65	0.00	0.06	69.33
Ours (pixel)	31.24	4.79	0.00	0.07	36.10
Ours (full)	31.51	5.34	20.29	0.10	57.24

Appendix

Table 11: Online runtime breakdown on PASLCD (RTX 5090; 20 instances, frames 1–25; ms/frame). Total sums all listed intervals. Post-frontend in Table 4 sums detection and refinement.

Scene	Inst.	Retained	Generated	Generation	Preprocess	Pixel	Contour	Total
		objects	records	(s) $\downarrow$	(s) $\downarrow$	(s) $\downarrow$	(s) $\downarrow$	(s) $\downarrow$
Cantina	1	22	366	106.95	8.21	41.68	18.84	175.68
	2	28	602	112.19	7.99	56.92	22.93	200.02
Garden	1	37	1080	119.42	7.57	83.23	30.48	240.69
	2	35	752	121.05	8.25	69.61	33.54	232.46
Lounge	1	18	384	112.07	8.26	42.39	19.14	181.86

Appendix

Table 12: Complete offline preparation and fitting runtime breakdown across 20 PASLCD scene-instances (seconds, measured on an NVIDIA RTX 5090). Generation includes automatic object proposals, 3DGS edits, and synthetic paired observation rendering. Generated records are counted before the camera split. Fitting covers reference preprocessing, pixel models, prior search, and contour filter fitting.

Figure 6: Qualitative comparison on PASLCD. Examples are from Cantina, Garden, Meeting room, and Porch. Each column shows one query, with reference, query, ground truth, and method outputs arranged vertically. Both O-SCD variants and our pixel and full outputs are shown. Green, orange, and pink denote correct detections, false positives, and missed changes, respectively. Numbers report per-frame foreground IoU. Thin structures remain partly missed in Garden.

Figure 7: Additional qualitative comparisons on PASLCD. Examples are from Printing area, Lounge, Lunch room, and another view of Cantina. The layout and colours follow Figure 6 . Refinement recovers the cabinet region in Printing area, while false positives near the ceiling in Lounge and missed object regions in Lunch room remain. Numbers report per-frame foreground IoU.

Figure 8: Qualitative comparison on CCS3D. Examples are from Livingroom, Desk, Bookcase, and Bedroom. Rows and colours follow Figure 6 . Ground truth is the union of pre- and post-state masks, as in Table 2 . Our full method reduces excess responses around displaced objects in Livingroom and Desk and on unchanged regions in Bedroom; boundary errors remain in Bookcase. Numbers report per-frame foreground IoU.

Figure 9: Evidence fusion in Cantina. Columns show frames 1, 13, 18, and 25 from one sequence. Log-odds maps share a colour scale saturated at $\pm 6$ for display; white denotes zero and grey marks uncovered pixels. The $C_{t}$ -only mask sets $H_{t}=\pi=0.5$ , giving $Q_{t}^{\mathrm{pixel}}=C_{t}$ . Both masks threshold $Q_{t}^{\mathrm{pixel}}$ at $0.5$ ; fusion combines the same $C_{t}$ , historical support $H_{t}$ from $F_{t-1}$ , and the fitted prior $\pi$ (Eq. 1 ). Mask colours follow Figure 6 ; numbers report per-frame foreground IoU.

Figure 10: Qualitative component ablations. Examples are from Cantina, Porch, Garden, and Meeting room in PASLCD. (a) Five pixel-pipeline ablations and the full pixel pipeline. The no-prior variant uses $\pi=0.5$ ; spatial support denotes Gaussian-aggregated support $S_{t}$ . (b) Three output-refinement ablations and the full method. The image-weight ablation omits image weighting during contour fitting. Both groups use the same queries. Colours follow Figure 6 ; numbers report per-frame foreground IoU.

3D Scene Change Detection

See all 13 papers

O-SCD++: Replay-Free Online Scene Change Detection via Recursive Evidence Fusion
Oct 17, 2026Chamuditha Jayanga Galappaththige, Jason Lai, Lloyd Windrim +43D Scene Change Detection 3D Scene Understanding
QUT Centre for Robotics, Queensland University of Technology, Australia · Australian Research Council Research Hub in Intelligent Robotic Systems for Real-Time Asset Management (ARIAM) · Australian Centre for Robotics, The University of Sydney, Australia +1
From Pixels to Primitives: Scene Change Detection in 3D Gaussian Splatting
May 8, 2026Chamuditha Jayanga Galappaththige, Jason Lai, Timothy Patten +33D Gaussian Splatting 3D Scene Change Detection
1QUT Centre for Robotics · 2ARIAM · 3ACFR, University of Sydney +1
GS-Pool: Object-Level Change Detection in 3D Gaussian Splatting
Oct 5, 2026Boaz Keren-Gil, James Gain, Patrick Marais3D Gaussian Splatting 3D Scene Change Detection
University of Cape Town

Oct 17, 2026J/K move · Enter open · S save

O-SCD++: Replay-Free Online Scene Change Detection via Recursive Evidence Fusion

Chamuditha Jayanga Galappaththige, Jason Lai, Lloyd Windrim, Timothy Patten+3

QUT Centre for Robotics, Queensland University of Technology, Australia · Australian Research Council Research Hub in Intelligent Robotic Systems for Real-Time Asset Management (ARIAM) · Australian Centre for Robotics, The University of Sydney, Australia+1

3D Scene Change Detection 3D Scene Understanding

Open paper AskPDF arXiv

Online scene change detection requires an agent to detect what has changed in a previously mapped environment while it is still on site, observing from unconstrained viewpoints and without access to future observations. This causal setting is strictly harder than its offline counterpart, and online methods have consistently trailed offline accuracy. O-SCD closed that gap by treating change as a persistent property of a 3D Gaussian splatting map of the reference scene, surpassing the strongest offline methods in change detection accuracy while operating at over 10 frames per second. O-SCD optimizes a change field over the Gaussians of the reference map, refitting it to all past cue maps at every frame, a step that consumes more than half of the per-frame budget and forces accuracy to be traded against speed. We present a replay-free formulation that eliminates this trade-off. Because the reference map stays fixed, rendering the change field is a linear operation. Each new observation can therefore be carried onto the map's Gaussians directly, by a single backward pass of the renderer, which reduces multi-view fusion to an exact two-scalar recursion per Gaussian. Past observations are never stored or revisited. The self-supervised objective that O-SCD minimizes by descent also determines the decision threshold exactly, by sorting, yielding a label-free per-frame cutoff whose margin provides a selective-prediction signal. On PASLCD, our method improves mean intersection over union from 0.489 to 0.575 and throughput from 11.8 to 25.6 frames per second, even exceeding O-SCD's offline result while remaining strictly causal.

Abstract

3D Scene Change Detection

Explore similar work

O-SCD++: Replay-Free Online Scene Change Detection via Recursive Evidence Fusion

From Pixels to Primitives: Scene Change Detection in 3D Gaussian Splatting

GS-Pool: Object-Level Change Detection in 3D Gaussian Splatting