High Dynamic Range Imaging Based on an Asymmetric Event-SVE Camera System

Pengju Sun; Banglei Guan; Jing Tao; Zhenbao Yu; Xuanyu Bai; Yang Shang; Qifeng Yu

cs.CVFeb 28, 2026

High Dynamic Range Imaging Based on an Asymmetric Event-SVE Camera System

Authors: Pengju Sun, Banglei Guan, Jing Tao, Zhenbao Yu, Xuanyu Bai, Yang Shang, Qifeng Yu

Organizations: College of Aerospace Science and Engineering, National University of Defense Technology, Hunan 410073, China · Hunan Provincial Key Laboratory of Image Measurement and Vision Navigation, Hunan 410073, China

Abstract

High dynamic range (HDR) imaging under extreme illumination remains challenging for conventional cameras due to overexposure. Event cameras provide microsecond temporal resolution and high dynamic range, while spatially varying exposure (SVE) sensors offer single-shot radiometric diversity.We present a hardware--algorithm co-designed HDR imaging system that tightly integrates an SVE micro-attenuation camera with an event sensor in an asymmetric dual-modality configuration. To handle non-coaxial geometry and heterogeneous optics, we develop a two-stage cross-modal alignment framework that combines feature-guided coarse homography estimation with a multi-scale refinement module based on spatial pooling and frequency-domain filtering. On top of aligned representations, we develop a cross-modal HDR reconstruction network with convolutional fusion, mutual-information regularization, and a learnable fusion loss that adaptively balances intensity cues and event-derived structural constraints. Comprehensive experiments on both synthetic benchmarks and real captures demonstrate that the proposed system consistently improves highlight recovery, edge fidelity, and robustness compared with frame-only or event-only HDR pipelines. The results indicate that jointly optimizing optical design, cross-modal alignment, and computational fusion provides an effective foundation for reliable HDR perception in highly dynamic and radiometrically challenging environments.

Figures & tables

Figure 1: Overview of event-assisted HDR reconstruction. Top: a conventional event+frame HDR pipeline typically fuses an intensity frame with an event representation using fixed fusion loss. Bottom: our asymmetric Event-SVE HDR system encodes multi-exposure radiometric measurements and performs calibration-guided coarse alignment followed by learnable refinement due to the non-coaxial setup.

Figure 2: Hybrid Event-SVE imaging platform. (a) Acquisition architecture in which a programmable trigger generator synchronizes the SVE camera’s exposure cycles with the event sensor’s asynchronous readout. (b) Temporal sampling characteristics showing discrete multi-exposure frame capture versus continuous event generation. (c) Hardware prototype with independent optical paths and a custom synchronization controller that provides a shared trigger reference for event-frame pairing.

Figure 3: Overview of the proposed cross-modal alignment and HDR fusion network. Multi-exposure SVE frames and event streams are encoded into multi-scale feature pyramids. At each pyramid level, the features are refined through spatial pooling and frequency-domain convolution to achieve cross-modal alignment. The aligned features are then aggregated and passed through an encoder–decoder sub-network to reconstruct the final HDR image.

Figure 4: Structure diagram of the FDconv module. FDConv transforms features to the frequency domain, applies a learnable spectral response via element-wise complex multiplication, and transforms back.

Algorithm 1: Learnable fusion loss (per iteration).

Input: SVE frame

M

, event representation

E

, SVE/event features

F_{\mathrm{SVE}},F_{\mathrm{evt}}

Output: Total loss

L_{\mathrm{total}}

and updated parameters

(\Theta,\theta)

\hat{I}\leftarrow\mathrm{Net}_{\Theta}(M,E)

(w_{\mathrm{exp}},w_{\mathrm{evt}})\leftarrow\mathrm{softmax}\!\big(G_{\theta}([F_{\mathrm{SVE}},F_{\mathrm{evt}}])\big)

3. Compute

L_{\mathrm{int}}

and

L_{\mathrm{grad}}

L_{\mathrm{fuse}}\leftarrow L_{\mathrm{int}}+\alpha\,L_{\mathrm{grad}}

Table 5

Figure 5: Qualitative comparison on the SVE-HDR dataset. Zoom in for better visualization of details. Our results provide more details and visual effects than other methods.

Method	E2VID [ 49 ]	Liu et al. [ 13 ]	Han et al. [ 50 ]	Li et al. [ 52 ]	HDRev-Net [ 2 ]	Ours
PSNR $\uparrow$	13.734	23.159	20.697	20.673	24.071	24.241
SSIM $\uparrow$	0.589	0.901	0.861	0.890	0.928	0.935
LPIPS $\downarrow$	0.451	0.104	0.208	0.151	0.110	0.102

Table 1: Quantitative comparison on the synthetic HDR dataset. All methods are evaluated on the same test split with synchronized event windows and SVE-like multi-exposure inputs. We report PSNR/SSIM (

\uparrow

) and LPIPS (

\downarrow

); best and second-best are highlighted.

Metric	SPD-MEF [ 53 ]	HDRev-Net [ 2 ]	E2VID+ [ 49 ]	EvLight [ 34 ]	Retinexformer [ 21 ]	Ours
NIQE $\downarrow$	8.33	9.43	16.44	14.38	17.29	10.97
PIQE $\downarrow$	14.18	22.81	30.06	16.16	35.67	12.82
Entropy $\uparrow$	5.33	4.65	6.47	5.27	6.30	6.91

Table 2: No-reference quality assessment on real captures. Lower (higher) is better for NIQE/PIQE (Entropy).

Figure 6: Real-world HDR reconstruction results captured by the proposed Event-SVE rig. Compared to frame-based methods, which suffer from overexposure in bright regions, and event-based approaches, which lack background context, our method provides more comprehensive information. Moreover, unlike other event-frame fusion techniques, our approach preserves clearer textures in both overexposed and background regions, delivering more useful visual details.

Removed Component	NIQE $\downarrow$	PIQE $\downarrow$	Entropy $\uparrow$	Notes
No alignment	13.81	34.24	5.52	Blur ghosting
No learnable fusion loss	10.94	26.32	6.31	reduced highlight recovery
Full system (Ours)	10.91	25.80	6.91	best

Table 3: Ablation study on major system components using no-reference metrics. Arrows indicate whether lower (

\downarrow

) or higher (

\uparrow

) is better.

HDR Imaging

High dynamic range imaging captures or reconstructs scene brightness beyond standard images, preserving detail in shadows and highlights.

See all 47 papers

Double-stream registration with pyramid fusion for HDR video with alternating exposures
Sep 25, 2026Onofre Martorell, Ivan Pereira-Sánchez, Antoni Fuentes +1HDR Video Reconstruction HDR Imaging
Dept. of Mathematics and Computer Science, Universitat de les Illes Balears, Cra. de Valldemossa km 7.5, Palma, 07122, Illes Balears, Spain
SEE Challenge 2026: Event-Guided Brightness Adjustment Across a Broad Illumination Range
Sep 24, 2026Yunfan Lu, Mingchao Xu, Hanyu Zhou +9Image Enhancement HDR Imaging
Hong Kong University of Science and Technology (Guangzhou) · National University of Singapore · Xidian University +5
High Dynamic Range Video Reconstruction from Single-Exposure Raw Sequences
Sep 23, 2026Tao Zhang, Peixian Su, Xingyu Gao +6HDR Imaging HDR Video Reconstruction
School of Communication Engineering, Hangzhou Dianzi University, Hangzhou, China, and also with Zhejiang Provincial Key Laboratory of Low Altitude Ubiquitous Networking Technology, Hangzhou Dianzi University, Hangzhou, China · Institute of Microelectronics, Chinese Academy of Sciences, Beijing, China, and also with the University of Chinese Academy of Sciences, Beijing, China · School of Computer Science and Technology, Beijing Institute of Technology, Beijing, China +1

Sep 25, 2026J/K move · Enter open · S save

Double-stream registration with pyramid fusion for HDR video with alternating exposures

Onofre Martorell, Ivan Pereira-Sánchez, Antoni Fuentes, Antoni Buades

Dept. of Mathematics and Computer Science, Universitat de les Illes Balears, Cra. de Valldemossa km 7.5, Palma, 07122, Illes Balears, Spain

HDR Video Reconstruction HDR Imaging Multi-Exposure Image Fusion Video Restoration

Open paper AskPDF arXiv

High dynamic range (HDR) video reconstruction from al-ter-na-ting-exposure sequences remains challenging, especially in regions with extreme luminance variation. We propose a novel HDR reconstruction framework based on dual-stream registration and accurate pyramid fusion. Given three consecutive frames, our method computes optical flow directly with the central frame, while introducing a complementary midpoint displacement strategy to handle cases with severe overexposition. A pyramid fusion stage then merges the resulting radiance and LDR images into a final HDR output. Experimental results demonstrate that our approach consistently outperforms state-of-the-art methods.

Abstract

HDR Imaging

Explore similar work

Double-stream registration with pyramid fusion for HDR video with alternating exposures

SEE Challenge 2026: Event-Guided Brightness Adjustment Across a Broad Illumination Range

High Dynamic Range Video Reconstruction from Single-Exposure Raw Sequences