F2HDR: Two-Stage HDR Video Reconstruction via Flow Adapter and Physical Motion Modeling

Huanjing Yue; Dawei Li; Shaoxiong Tu; Jingyu Yang

cs.CVMar 16, 2026

F2HDR: Two-Stage HDR Video Reconstruction via Flow Adapter and Physical Motion Modeling

Authors: Huanjing Yue, Dawei Li, Shaoxiong Tu, Jingyu Yang

Organizations: School of Electrical and Information Engineering, Tianjin University · State Key Laboratory of Smart Power Distribution Equipment and System, Tianjin University · Huawei Technologies Co., Ltd.

Abstract

Reconstructing High Dynamic Range (HDR) videos from sequences of alternating-exposure Low Dynamic Range (LDR) frames remains highly challenging, especially under dynamic scenes where cross-exposure inconsistencies and complex motion make inter-frame alignment difficult, leading to ghosting and detail loss. Existing methods often suffer from inaccurate alignment, suboptimal feature aggregation, and degraded reconstruction quality in motion-dominated regions. To address these challenges, we propose F2HDR, a two-stage HDR video reconstruction framework that robustly perceives inter-frame motion and restores fine details in complex dynamic scenarios. The proposed framework integrates a flow adapter that adapts generic optical flow for robust cross-exposure alignment, a physical motion modeling to identify salient motion regions, and a motion-aware refinement network that aggregates complementary information while removing ghosting and noise. Extensive experiments demonstrate that F2HDR achieves state-of-the-art performance on real-world HDR video benchmarks, producing ghost-free and high-fidelity results under large motion and exposure variations.

Figures & tables

Figure 1: Comparison on the real-world HDR video reconstruction benchmark [ 30 ] with state-of-the-art methods [ 35 , 6 ] . The challenging scene involves both global and local motion, while the reference frames suffer from severe information loss. Our $\text{F}^{2}\text{HDR}$ achieves the most robust optical flow estimation and the most accurate detail restoration.

Figure 2: Network architecture of the proposed $\text{F}^{2}\text{HDR}$ , which includes coarse fusion stage and refinement stage.

Figure 3: Qualitative comparisons with state-of-the-art methods on DeepHDRVideo dataset [ 2 ] .

Figure 4: Qualitative comparisons with state-of-the-art methods on Real-HDRV dataset [ 30 ] .

Methods		PSNR $T$	SSIM $T$	HDR-VDP-2
Kalantari13 [ 14 ]	TOG2013	40.33	0.9409	–
AHDRNet [ 37 ]	CVPR2019	40.54	0.9452
Prabhakar [ 28 ]	CVPR2021	40.21	0.9414
HDR-Transformer [ 21 ]	ECCV2022	41.51	0.9458
Kalantari19 [ 13 ]	CGF2019	39.91	0.9329	–
Chen21 [ 2 ]	ICCV2021	43.32	0.9551	78.37

Table 1: Quantitative comparisons on the DeepHDRVideo dataset [ 2 ] , including both image-based [ 14 , 37 , 28 , 21 ] and video-based methods [ 13 , 2 , 5 , 6 , 35 , 8 ] . * indicates additional training with the Sintel dataset. Red Bold : best, Blue Bold : second best.

Methods	PSNR $T$	SSIM $T$	HDR-VDP-2	PSNR $L$	SSIM $L$
Chen21 [ 2 ]	40.79	0.9510	76.50	50.30	0.9943
LAN-HDR [ 5 ]	39.79	0.9448	74.43	46.54	0.9928
HDRFlow [ 35 ]	40.34	0.9481	75.82	49.72	0.9941
HDRFlow* [ 35 ]	40.28	0.9450	76.08	49.87	0.9941
NECHDR [ 6 ]	40.88	0.9518	76.77	50.41	0.9952
Ours	41.01	0.9538	76.93	50.51	0.9949

Table 2: Quantitative comparison of different methods on the Real-HDRV dataset [ 30 ] . * indicates additional training with the Sintel dataset. Red Bold : best, Blue Bold : second best.

	HDRFlow	NECHDR	Ours w/o FA	Ours
Optical Flow EPE $\downarrow$	3.53	3.30	1.81	2.08
LDR Warping PSNR $\uparrow$	29.93	30.31	30.39	30.84
LDR Warping SSIM $\uparrow$	0.8763	0.8901	0.8897	0.8922

Table 3: Comparison of optical flow accuracy on Real-HDRV [ 30 ] .

Figure 5: Qualitative Comparisons of optical flow and warped LDR on DeepHDRVideo (Rows 1-2) [ 2 ] and Real-HDRV (Rows 3-4) [ 30 ] Datasets.

Methods	Time (s)
Methods	1920×1080	1536×813
Kalantari19 [ 13 ]	0.260	0.220
HDRFlow [ 35 ]	0.076	0.050
NECHDR [ 6 ]	0.160	0.103
Chen21 [ 2 ]	0.570	0.540
LAN-HDR [ 5 ]	0.905	0.525

Table 4: Inference time comparison for two different resolutions evaluated on an NVIDIA RTX 3090 GPU.

Methods	DeepHDRVideo		Real-HDRV
Methods	PSNR $T$	SSIM $T$	PSNR $T$	SSIM $T$
Stage I w/o FA	43.06	0.9514	40.29	0.9453
Stage I w/o FA + Stage II	43.21	0.9530	40.45	0.9461
Stage I	43.39	0.9549	40.61	0.9481
Stage I w/o Mask + Stage II	43.68	0.9565	40.87	0.9502
Stage I w/ Mask* + Stage II	43.76	0.9569	40.93	0.9524

Table 5: Ablation study of

\text{F}^{2}\text{HDR}

on DeepHDRVideo dataset and Real-HDRV dataset. FA denotes Flow Adapter. Mask* denotes the mask learned via a CNN.

Figure 6: Ablation of flow adapter by removing it from our full solution.

HDR Video Reconstruction

See all 9 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
LoCAtion: Long-time Collaborative Attention Framework for High Dynamic Range Video Reconstruction
Mar 15, 2026Qianyu Zhang, Bolun Zheng, Lingyu Zhu +3HDR Video Reconstruction HDR Imaging
School of Automation, Hangzhou Dianzi University, Hangzhou 310018, China · Department of Computer Science, City University of Hong Kong
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 Video Reconstruction

Explore similar work

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

LoCAtion: Long-time Collaborative Attention Framework for High Dynamic Range Video Reconstruction

High Dynamic Range Video Reconstruction from Single-Exposure Raw Sequences