RSEdit: Text-Guided Image Editing for Remote Sensing

Chen Zhenyuan; Zhang Zechuan; Zhang Feng

cs.CVMar 14, 2026

RSEdit: Text-Guided Image Editing for Remote Sensing

Authors: Chen Zhenyuan, Zhang Zechuan, Zhang Feng

Organizations: School of Earth Sciences, Zhejiang University, Hangzhou, 310058 China, and Zhejiang Provincial Key Laboratory of Geographic Information Science. · ReLER, CCAI, Zhejiang University, Hangzhou, China. · Key Laboratory of Spatio-temporal Information and Intelligent Services (LSIIS), Ministry of Natural Resources of the People’s Republic of China.

Abstract

In this paper, we explore text-guided image editing in the remote sensing domain using generative modeling. We propose \rsedit, a collection of models from U-Net to DiT with various configurations. Specifically, we present the first comprehensive study of conditioning strategies for building image editing models from off-the-shelf text-to-image ones. Our experiments show that \rsedit achieves the best instruction-faithful edits while preserving geospatial structure. We release the code at https://github.com/Bili-Sakura/RSEdit-Preview and checkpoints at https://huggingface.co/collections/BiliSakura/rsedit.

Figures & tables

Fig. 1: Illustration of text-guided remote sensing image editing with RSEdit .

Fig. 2: Architecture-aware conditioning in RSEdit . U-Net: (a) channel concatenation, (b) token concatenation. DiT: (c) conditioning with token layout $(2N,d)$ ; (d) conditioning with token layout $(N,2d)$ . Curves denote the DiT token flatten order, which is handled by 2D position embedding.

Fig. 3: Receptive field comparison: (a) convolutional kernel, (b) full self-attention, (c) window attention. Shaded regions indicate context seen from the query (star). Window attention balances the compute budget and receptive-field extent between these extremes.

Fig. 4: Architecture of the RSEdit DiT+ variant.

Fig. 5: Qualitative comparison on disaster simulation. From left: Edit prompt, Input, Reference, RSEdit U-Net , RSEdit DiT , InstructPix2Pix, UltraEdit. RSEdit produces physically plausible edits while baselines under-edit or introduce artifacts.

Method	F1 dam $\uparrow$	VIE Scores $\uparrow$
Method		SC	PQ	VIE
SD 1.5 †	$6.57$	$1.680$	$5.104$	$2.345$
SD 2.1 †	$7.90$	$1.672$	$4.479$	$2.148$
Text2Earth †	$15.28$	$2.391$	$5.003$	$2.922$
InstructPix2Pix	$8.35$	$3.435$	$\underline{5.219}$	$3.305$
MagicBrush	$0.94$	$1.145$	$4.642$	$1.647$

TABLE I: Change detection evaluation results on RSCC.

F1_{\text{dam}}

follows Eq. ( 3 ) (xView2-style aggregation). Bold indicates best, underline second best.

Fig. 6: Proposed change-centric evaluation protocol. $F1_{\text{dam}}$ measures localized change correctness.

	U-Net		DiT		DiT+
Metric	ch.	tk.	tk.	ch.	tk.
#Param	$\sim$ 860M	$\sim$ 860M	$\sim$ 610M	$\sim$ 610M	$\sim$ 610M
GFLOPs	$\sim$ 1 $\times$	$\sim$ 2 $\times$	$\sim$ 3 $\times$	$\sim$ 1.5 $\times$	$\sim$ 2.5 $\times$
F1 dam $\uparrow$	49.74	38.29	30.28	41.37	45.35
SC $\uparrow$	4.050	4.024	4.509	4.337	3.604
PQ $\uparrow$	5.358	5.257	5.219	5.429	5.160

TABLE II: Ablation on conditioning strategy across U-Net and DiT backbones on RSCC. Columns are backbone/fusion pairs: U-Net with channel (ch.) or token (tk.) concatenation; DiT with token or channel fusion; DiT+ denotes RSEdit DiT+ with token fusion.

Fig. 7: Additional qualitative comparison (six exemplar rows). Columns left to right: pre-event (GT), post-event (GT), GT color-coded change mask, RSEdit U-Net with channel concatenation, RSEdit U-Net with token concatenation, RSEdit DiT with token concatenation, RSEdit DiT with channel concatenation, and RSEdit DiT+ (token concatenation).

Text-Guided Image Editing

Text-guided image editing changes selected image content from natural-language instructions while aiming to preserve everything else.

See all 165 papers

RS-RIE-Bench: Benchmarking Reasoning-Guided Remote Sensing Image Editing
Jul 22, 2026Zihan Qin, Boao Xu, Zhao Dong +4Image Editing Evaluation Remote Sensing Benchmarks
School of Artificial Intelligence, Optics and Electronics (iOPEN), Northwestern Polytechnical University, Xi’an 710072, China. · Xi’an Modern Control Technology Research Institute, Xi’an 710065, China. · Lanzhou Institute of Physics, Lanzhou 730000, China. +2
Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network
Sep 17, 2026Yulong Chen, Ziqian Zhang, Haoyu Zhang +4Text-Guided Image Editing Training-Free Image Editing
City University of Hong Kong (Dongguan), Guangdong, China · City University of Hong Kong, Hong Kong, China · Mohamed bin Zayed University of Artificial Intelligence, Masdar, Abu Dhabi
Prompt-Guided Image Editing with Masked Logit Nudging in Visual Autoregressive Models
Apr 16, 2026Amir El-Ghoussani, Marc Hölle, Gustavo Carneiro +1Text-Guided Image Editing Visual Autoregressive Models
Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany · University of Surrey, United Kingdom

Jul 22, 2026J/K move · Enter open · S save

RS-RIE-Bench: Benchmarking Reasoning-Guided Remote Sensing Image Editing

Zihan Qin, Boao Xu, Zhao Dong, Yingping Sun+3

School of Artificial Intelligence, Optics and Electronics (iOPEN), Northwestern Polytechnical University, Xi’an 710072, China. · Xi’an Modern Control Technology Research Institute, Xi’an 710065, China. · Lanzhou Institute of Physics, Lanzhou 730000, China.+2

Image Editing Evaluation Remote Sensing Benchmarks Image Editing

Open paper AskPDF arXiv

Remote sensing image editing aims to modify remote sensing images according to natural language instructions while preserving geographic rules and sensor observation characteristics. Existing benchmarks mainly target natural images or general visual scenes, and thus may not fully capture the reasoning, regional control, and sensor-consistency abilities required in remote sensing editing. To fill this gap, we introduce RS-RIE-Bench, the first benchmark for reasoning-guided remote sensing image editing. RS-RIE-Bench organizes tasks into three categories: temporal reasoning, causal reasoning, and spatial reasoning. These categories capture temporal evolution, causal consequence, and spatial imaging consistency in remote sensing scenes. The evaluation protocol covers three dimensions: target region plausibility, non-target region preservation, and image quality consistency. We further demonstrate the feasibility of MLLM-based evaluation through cross-judge consistency analysis and stratified expert review. Systematic evaluation on eight open-source and closed-source image editing models shows that current models still have clear limitations in reasoning-guided remote sensing editing. Even the strongest model achieves only 24.28% overall accuracy under the strict joint-satisfaction criterion, while the mean relaxed joint-4 success rate across all eight models is 32.23%. Causal reasoning and spatial reasoning remain especially challenging, and several open-source models are close to zero in some categories. These results show that RS-RIE-Bench can effectively reveal the limitations of current models in geographic reasoning, regional control, and sensor-consistent generation. It also provides a standardized benchmark and a clear research direction for future remote sensing intelligent editing models.

Abstract

Text-Guided Image Editing

Explore similar work

RS-RIE-Bench: Benchmarking Reasoning-Guided Remote Sensing Image Editing

Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network

Prompt-Guided Image Editing with Masked Logit Nudging in Visual Autoregressive Models