ORSIFlow: Saliency-Guided Rectified Flow for Optical Remote Sensing Salient Object Detection

Haojing Chen; Zhihang Liu; Yutong Li; Tao Tan; Haoyu Bian; Qiuju Ma

cs.CVMar 30, 2026

ORSIFlow: Saliency-Guided Rectified Flow for Optical Remote Sensing Salient Object Detection

Authors: Haojing Chen, Zhihang Liu, Yutong Li, Tao Tan, Haoyu Bian, Qiuju Ma

Organizations: University of Electronic Science and Technology of China, Chengdu, China · Hainan University, Haikou, China · China University of Mining & Technology (Beijing), Beijing, China

Abstract

Optical Remote Sensing Image Salient Object Detection (ORSI-SOD) remains challenging due to complex backgrounds, low contrast, irregular object shapes, and large variations in object scale. Existing discriminative methods directly regress saliency maps, while recent diffusion-based generative approaches suffer from stochastic sampling and high computational cost. In this paper, we propose ORSIFlow, a saliency-guided rectified flow framework that reformulates ORSI-SOD as a deterministic latent flow generation problem. ORSIFlow performs saliency mask generation in a compact latent space constructed by a frozen variational autoencoder, enabling efficient inference with only a few steps. To enhance saliency awareness, we design a Salient Feature Discriminator for global semantic discrimination and a Salient Feature Calibrator for precise boundary refinement. Extensive experiments on multiple public benchmarks show that ORSIFlow achieves state-of-the-art performance with significantly improved efficiency.

Figures & tables

Fig. 1: The left panel shows performance in terms of $S_{\alpha}$ on three datasets, while the right panel compares our method against a previous state-of-the-art approach. Red boxes indicate segmentation errors.

Fig. 2: Overall architecture of the proposed method. The SFCN extracts multi-scale conditional features via SFD and SFC modules , while the LSFN learns a deterministic velocity field to transform the Gaussian distribution into saliency maps within the VAE latent space.

Method	Pub	ORSSD				EORSSD				ORSI-4199
Method	Pub	$S_{\alpha}\uparrow$	$F_{\beta}^{max}\uparrow$	$E_{\xi}^{max}\uparrow$	$MAE\downarrow$	$S_{\alpha}\uparrow$	$F_{\beta}^{max}\uparrow$	$E_{\xi}^{max}\uparrow$	$MAE\downarrow$	$S_{\alpha}\uparrow$	$F_{\beta}^{max}\uparrow$	$E_{\xi}^{max}\uparrow$	$MAE\downarrow$
GeleNet [ 13 ]	TIP-23	0.9328	0.9109	0.9760	0.0099	0.9324	0.8796	0.9735	0.0063	0.8730	0.8663	0.9397	0.0318
ACCoNet [ 15 ]	TCYB-23	0.9437	0.9149	0.9796	0.0080	0.9290	0.8837	0.9727	0.0074	0.8817	0.8740	0.9478	0.0294
ERPNet [ 26 ]	TCYB-23	0.9352	0.9036	0.9738	0.0114	0.9252	0.8743	0.9665	0.0082	0.8684	0.8610	0.9384	0.0327
WeightNet [ 4 ]	TGRS-24	0.9512	0.9278	0.9864	0.0075	0.9417	0.8908	0.9809	0.0060	0.8868	0.8821	0.9496	0.0290
PRNet [ 8 ]	TGRS-24	0.9425	0.9153	0.9866	0.0068	0.9296	0.8707	0.9809	0.0053	0.8795	0.8734	0.9491	0.0280

TABLE I: Quantitative comparison results in terms of

F_{\beta}^{max}

E_{\xi}^{max}

S_{\alpha}

and

MAE

score are presented on three benchmark datasets. In the tables,

\uparrow

and

\downarrow

indicate that higher and lower values are preferred, respectively. The top three results in each column are highlighted in red, blue, and green, respectively.

Fig. 3: Qualitative comparison of ORSIFlow and other state-of-the-art methods

NO.	Module Ablation			ORSSD (200 test images)				EORSSD (600 test images)				ORSI-4199 (2199 test images)
NO.	VAE	SFD	SFC	$S_{\alpha}\!\uparrow$	$F_{\beta}^{max}\!\uparrow$	$E_{\xi}^{max}\!\uparrow$	$MAE\!\downarrow$	$S_{\alpha}\!\uparrow$	$F_{\beta}^{max}\!\uparrow$	$E_{\xi}^{max}\!\uparrow$	$MAE\!\downarrow$	$S_{\alpha}\!\uparrow$	$F_{\beta}^{max}\!\uparrow$	$E_{\xi}^{max}\!\uparrow$	$MAE\!\downarrow$
(a)				0.9453	0.9003	0.9841	0.0063	0.9303	0.8846	0.9787	0.0051	0.8696	0.8593	0.9437	0.0302
(b)	✓			0.9479	0.9067	0.9847	0.0061	0.9326	0.8882	0.9794	0.0050	0.8740	0.8651	0.9461	0.0297
(c)		✓		0.9548	0.9189	0.9863	0.0056	0.9398	0.8970	0.9821	0.0046	0.8866	0.8789	0.9515	0.0274
(d)			✓	0.9516	0.9128	0.9856	0.0059	0.9362	0.8923	0.9811	0.0048	0.8801	0.8724	0.9492	0.0287
(e)	✓	✓		0.9567	0.9217	0.9875	0.0052	0.9415	0.8989	0.9831	0.0044	0.8894	0.8826	0.9530	0.0271

TABLE II: Comparison of different module ablation settings on ORSSD, EORSSD and ORSI-4199 datasets. The best result in each column is highlighted in bold.

Fig. 4: Visual results of the effectiveness of our modules.

Method	Timestep	FLOPs (G)	FPS	$S_{\alpha}\uparrow$	$F_{\beta}^{\max}\uparrow$	$E_{\xi}^{\max}\uparrow$	$MAE\downarrow$
CamoDiff	10	60.29	0.3	0.8772	0.8710	0.9432	0.0312
ORSIDiff	10	15.25	8	0.8860	0.8812	0.9510	0.0275
Ours	3	5.68	19	0.8935	0.8870	0.9545	0.0262

TABLE III: Efficiency comparison on the ORSI-4199 dataset with different time steps among generative ORSI-SOD methods. The best result in each column is highlighted in bold.

Explore similar work

IPDiff: Diffusion-driven ORSI Salient Object Detection with Information Reconstruction and Multi-Prior Guidance
Jul 4, 2026Gongyang Li, Zhen Bai, Runmin Cong +3Salient Object Detection Diffusion Guidance
School of Communication and Information Engineering, Shanghai University, Shanghai, China · Department of Medical Equipment, the First Affiliated Hospital of Zhengzhou University, Zhengzhou, China · School of Control Science and Engineering, Shandong University, Jinan, China +2
SPLG-Mamba: Structure-Preserving Local-Global Mamba Network for Salient Object Detection in Optical Remote Sensing Images
Aug 30, 2026Yi Xu, Ruichao Hou, Tongwei Ren +1Mamba Salient Object Detection
State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, 210008, Jiangsu, China · School of Elite Biomedical Engineers and the Institute for Interdisciplinary Intelligent Pharmacy, China Pharmaceutical University, Nanjing 211198, China
ChangeFlow -- Latent Rectified Flow for Change Detection in Remote Sensing
May 14, 2026Blaž Rolih, Matic Fučka, Filip Wolf +1Remote Sensing Change Detection
University of Ljubljana, Faculty of Computer and Information Science, Slovenia

Jul 4, 2026J/K move · Enter open · S save

IPDiff: Diffusion-driven ORSI Salient Object Detection with Information Reconstruction and Multi-Prior Guidance

Gongyang Li, Zhen Bai, Runmin Cong, Dan Zeng+2

School of Communication and Information Engineering, Shanghai University, Shanghai, China · Department of Medical Equipment, the First Affiliated Hospital of Zhengzhou University, Zhengzhou, China · School of Control Science and Engineering, Shandong University, Jinan, China+2

Salient Object Detection Diffusion Guidance Conditional Diffusion Models

Open paper AskPDF arXiv

Existing Salient Object Detection in Optical Remote Sensing Image (ORSI-SOD) methods mainly adopt the static inference strategy, which uses fixed trained model parameters for saliency inference in the testing phase. This means that even if the generated saliency map has errors, it cannot be further optimized. In this paper, we propose the novel IPDiff, a Diffusion-driven ORSI-SOD method with Information Reconstruction and Multi-Prior Guidance. We build IPDiff based on a unique dynamic optimization strategy, which endows IPDiff with the ability to iteratively optimize saliency maps with a dynamic parameter. Specifically, we formulate ORSI-SOD as a conditional diffusion problem in IPDiff. IPDiff first extracts informative conditional priors from ORSIs, including the saliency prior and the hierarchical priors, in the prior network with the assistance of the information reconstruction-driven attention module. The saliency prior can provide positional information of salient objects, while the hierarchical priors can provide specific detail and semantic information of salient objects. Under the guidance of these priors, IPDiff then iteratively denoises random noise as the timestep dynamically changes in the denoising network, generating saliency maps that are close to ground truths. Notably, we simultaneously supervise IPDiff in both spatial and spectral domains through a hybrid loss function to achieve efficient network training. Comprehensive experiments on public ORSSD, EORSSD, and ORSI-4199 datasets demonstrate that our proposed IPDiff achieves the best performance compared to 46 state-of-the-art methods. The code and results of our method are available at https://github.com/MathLee/IPDiff.

Abstract

Explore similar work

IPDiff: Diffusion-driven ORSI Salient Object Detection with Information Reconstruction and Multi-Prior Guidance

SPLG-Mamba: Structure-Preserving Local-Global Mamba Network for Salient Object Detection in Optical Remote Sensing Images

ChangeFlow -- Latent Rectified Flow for Change Detection in Remote Sensing