SFE-VGGT: Source-Free VGGT Distillation for Event-Based Monocular Depth Estimation

Thai Duy Nguyen; Addison Lin Wang

cs.CVSep 29, 2026

SFE-VGGT: Source-Free VGGT Distillation for Event-Based Monocular Depth Estimation

Authors: Thai Duy Nguyen, Addison Lin Wang

Organizations: Nanyang Technological University, Singapore.

Abstract

Recent event-based depth estimation methods successfully transfer geometric priors from vision foundation models via cross-modal distillation. However, their reliance on synchronized RGB-event pairs or depth annotations during training severely restricts practical deployment. To overcome this bottleneck, we propose SFE-VGGT, a novel source-free framework that distills the geometric priors of VGGT to the event domain without any paired RGB observations. Our core idea is to reconstruct surrogate frames directly from the target event stream to act as a frozen geometric teacher, entirely eliminating the need for genuine source RGB data. Crucially, as these surrogate frames inherently yield imperfect and spatially varying supervision, directly distilling from them propagates artifacts. To resolve this, we introduce a novel reliability-aware distillation strategy. This includes Density-Aware Feature Distillation to emphasize informative event regions, and Confidence-Weighted Depth Distillation to dynamically regulate supervision based on relative teacher-student prediction confidence. Meanwhile, we propose a Cross-Frame Relational Consistency loss that enforces temporal geometric stability using reliable inter-frame correspondences, bypassing the need for temporally consistent teacher's depth. Extensive experiments demonstrate that, despite source-free, our SFE-VGGT closely matches the accuracy of RGB-dependent baselines under standard conditions and significantly surpasses them in challenging nighttime scenarios. Across MVSEC nighttime sequences, SFE-VGGT reduces the average 10 m depth error by 15.3% compared with EventVGGT. Moreover, our method exhibits robust zero-shot generalization across real-world datasets, proving that highly effective geometric priors can be transferred to event cameras using strictly source-free supervision.

Figures & tables

Method	Training		Inference	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$
Method	Source-free	Depth GT	Input	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$
E2Depth [ 5 ]	–	✓	E	1.79	5.35	8.31
RAMNet [ 6 ]	–	✓	E+I	0.81	2.26	3.58
HMNet [ 25 ]	–	✓	E+I	0.55	1.80	3.27
ER-F2D [ 7 ]	–	✓	E+I	0.67	1.69	2.81
SRFNet [ 2 ]	–	✓	E+I	1.27	1.68	2.76

TABLE I: Mean absolute depth error on EventScape in meters. E denotes event-only input and E+I denotes event and RGB image input. “Depth GT” indicates whether metric ground-truth depth values are used as training supervision. Best and second-best results are highlighted.

Method	Training		Inference	Night1			Night2			Night3			Day1
Method	Source-free	Depth GT	Input	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$
E2Depth [ 5 ]	–	✓	E	3.38	3.82	4.46	1.67	2.63	3.58	1.42	2.33	3.18	1.67	2.64	3.13
RAMNet [ 6 ]	–	✓	E+I	2.50	3.19	3.82	1.21	2.31	3.28	1.01	2.34	3.43	1.39	2.17	2.76
EvT+ [ 22 ]	–	✓	E+I	1.45	2.10	2.88	1.48	2.13	2.90	1.38	2.03	2.77	1.24	1.91	2.36
HMNet [ 25 ]	–	✓	E+I	1.50	2.48	3.19	1.36	2.25	2.96	1.27	2.17	2.86	1.22	2.21	2.68
ER-F2D [ 7 ]	–	✓	E+I	1.58	2.24	2.78	1.54	2.23	2.95	1.24	1.96	2.81	1.34	2.25	2.62

TABLE II: Mean absolute depth error on MVSEC in meters. E denotes event-only input and E+I denotes event and RGB image input. “Depth GT” indicates whether metric ground-truth depth values are used as training supervision. Among distillation-based methods, the Best and second-best results are highlighted. Overall best results across all methods are shown in bold .

Fig. 3: Qualitative results on MVSEC dataset.

Method	Input	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$
RAMNet [ 6 ]	E+I	2.62	11.26	19.11
SRFNet [ 2 ]	E+I	1.50	3.57	6.12
EventDAM [ 4 ]	E	1.20	2.60	5.18
EventVGGT [ 8 ]	E	0.54	0.89	1.33
SFE-VGGT	E	1.04	1.44	2.35

TABLE III: Zero-shot depth estimation on DENSE. Mean absolute depth error is reported in meters.

Method	Input	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$	$\delta_{1}$ $\uparrow$	$\delta_{2}$ $\uparrow$	$\delta_{3}$ $\uparrow$
VGGT [ 9 ]	E	2.42	2.71	3.44	0.42	0.69	0.84
VGGT [ 9 ]	I	2.31	2.68	3.33	0.41	0.71	0.88
EventVGGT [ 8 ]	E	1.67	2.02	2.61	0.57	0.82	0.93
SFE-VGGT	E	1.41	2.04	2.92	0.48	0.71	0.83

TABLE IV: Direct VGGT transfer analysis on MVSEC Night1. Mean absolute depth error is reported in meters.

$\mathcal{L}_{DAFD}$	$\mathcal{L}_{CWDD}$	$\mathcal{L}_{CFRC}$	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$	$\delta_{1}$ $\uparrow$	$\delta_{2}$ $\uparrow$	$\delta_{3}$ $\uparrow$
	$\checkmark$		0.620	0.951	1.337	0.740	0.880	0.948
$\checkmark$	$\checkmark$		0.622	0.947	1.327	0.739	0.875	0.942
	$\checkmark$	$\checkmark$	0.604	0.948	1.336	0.752	0.887	0.947
$\checkmark$	$\checkmark$	$\checkmark$	0.617	0.941	1.322	0.743	0.880	0.948

TABLE V: Ablation of the proposed training objectives.

Settings	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$
w/o event-density weighting ( $\mathcal{L}_{DAFD}$ )	0.633	0.965	1.347
w/o relative-confidence weighting ( $\mathcal{L}_{CWDD}$ )	0.631	0.954	1.334
w/o RANSAC ( $\mathcal{L}_{CFRC}$ )	0.630	0.944	1.325
Full model (all activated)	0.617	0.941	1.322

TABLE VI: Ablation of key components within the proposed objectives.

Frames	10m $\downarrow$	20m $\downarrow$	30m $\downarrow$	$\delta_{1}$ $\uparrow$	$\delta_{2}$ $\uparrow$	$\delta_{3}$ $\uparrow$
3	0.666	1.034	1.510	0.693	0.834	0.911
6	0.681	1.025	1.471	0.706	0.847	0.921
12	0.677	1.007	1.424	0.725	0.866	0.937
24	0.617	0.941	1.322	0.743	0.880	0.943

TABLE VII: Effect of input sequence length.

Explore similar work

Mar 10, 2026cs.CV

EventVGGT: Exploring Cross-Modal Distillation for Consistent Event-based Depth Estimation

Event cameras offer superior sensitivity to high-speed motion and extreme lighting, making event-based monocular depth estimation a promising approach for robust 3D perception in challenging conditions. However, progress is severely hindered by the scarcity of dense depth annotations. While recent annotation-free approaches mitigate this by distilling knowledge from Vision Foundation Models (VFMs), a critical limitation persists: they process event streams as independent frames. By neglecting the inherent temporal continuity of event data, these methods fail to leverage the rich temporal priors encoded in VFMs, ultimately yielding temporally inconsistent and less accurate depth predictions. To address this, we introduce EventVGGT, a novel framework that explicitly models the event stream as a coherent video sequence. To the best of our knowledge, we are the first to distill spatio-temporal and multi-view geometric priors from the Visual Geometry Grounded Transformer (VGGT) into the event domain. We achieve this via a comprehensive tri-level distillation strategy: (i) Cross-Modal Feature Mixture (CMFM) bridges the modality gap at the output level by fusing RGB and event features to generate auxiliary depth predictions; (ii) Spatio-Temporal Feature Distillation (STFD) distills VGGT's powerful spatio-temporal representations at the feature level; and (iii) Temporal Consistency Distillation (TCD) enforces cross-frame coherence at the temporal level by aligning inter-frame depth changes. Extensive experiments demonstrate that EventVGGT consistently outperforms existing methods -- reducing the absolute mean depth error at 30m by over 53% on EventScape (from 2.30 to 1.06) -- while exhibiting robust zero-shot generalization on the unseen DENSE and MVSEC datasets. The code is available at https://github.com/yinruiRen/EventVGGT.

Yinrui Ren, Jinjing Zhu, Kanghao Chen +8

Hong Kong University of Science and Technology (Guangzhou) · South China Normal University · Chinese University of Hong Kong

Monocular Depth Estimation Depth Estimation Cross-Modal Distillation Dense Annotation

Ask PDF

May 11, 2026cs.CV

Neuromorphic Monocular Depth Estimation with Uncertainty Modeling

Event cameras offer distinct advantages over conventional frame-based sensors, including microsecond-level temporal resolution, high dynamic range, and low bandwidth. In this paper, we predict per-pixel depth distributions from monocular event streams using deep neural networks. We estimate uncertainty using Gaussian, log-normal, and evidential learning frameworks. We compare six event representations: spatio-temporal voxel grids with 1, 5, 10, and 20 temporal bins, the Compact Spatio-Temporal Representation (CSTR), and Time-Ordered Recent Event (TORE) volumes. Our U-Net-based models are trained on synthetic data and then fine-tuned on real sequences. We evaluate performance using absolute relative error, root mean squared error, and the area under the sparsification error. Quantitative results show that the representations perform similarly, while 10 bin log-normal and 5 bin evidential learning perform best across metrics. Our experiments demonstrate that uncertainty estimation can be successfully integrated into event-based monocular depth estimation, and be used to indicate pixels with reliable depth.

Viktor Bergkvist, Felix Rydell, Per-Erik Forssén +2

Swedish Defence Research Agency Linköping, Sweden · Swedish Defence Research Agency Kista, Sweden · Linköping University Linköping, Sweden

Monocular Depth Estimation Monocular

Ask PDF

Jul 17, 2026cs.CV

Geometric Distillation from Rectified Stereo: Leveraging Epipolar Cues for Monocular Depth

Monocular depth foundation models have demonstrated remarkable generalization capabilities across diverse environments. However, they continue to struggle with metric depth estimation in diverse environments. This limitation stems from the inherent scale ambiguity of single-view inference, leading to misaligned scale predictions even when the relative geometry is accurate. Conversely, recent multi-view foundation models leverage cross-view cues to learn robust scene-level geometry and consistent scale. Yet, these benefits typically vanish during single-image inference, as the absence of explicit geometric constraints causes performance to degrade. To bridge this gap, we propose a novel framework that transfers the scale-aware geometric priors of multi-view models into monocular depth foundation models. Specifically, we introduce an Epipolar Distillation (EpiDistill), an approach utilizing Rectified Stereo Tokens, which enables the single-view prediction model to retain epipolar attention patterns and maintain geometric consistency without requiring multi-view inputs at inference. Experimental results demonstrate that our method significantly improves zero-shot metric depth estimation, particularly on challenging datasets like ETH3D and DIODE where scale alignment is critical. Furthermore, our approach is model-agnostic, consistently boosting the performance of state-of-the-art ViT-based models, including UniDepthV2 and DepthPro.

Jung-Hee Kim, Xiaoming Liu

Michigan State University, East Lansing, MI 48824, USA · University of North Carolina at Chapel Hill, Chapel Hill, NC 27514, USA

Monocular Depth Estimation Scale-Ambiguous Epipolar Geometry Stereo Vision Robust Geometric Model Estimation Geometric Priors Cross-View+2

Ask PDF