EarthBridge: A Solution for 4th Multi-modal Aerial View Image Challenge Translation Track

Zhenyuan Chen; Guanyuan Shen; Feng Zhang

cs.CVMar 6, 2026

EarthBridge: A Solution for 4th Multi-modal Aerial View Image Challenge Translation Track

Authors: Zhenyuan Chen, Guanyuan Shen, Feng Zhang

Organizations: Zhejiang University · University of Science and Technology of China

Abstract

Cross-modal image-to-image translation among Electro-Optical (EO), Infrared (IR), and Synthetic Aperture Radar (SAR) sensors is essential for comprehensive multi-modal aerial-view analysis. However, translating between these modalities is notoriously difficult due to their distinct electromagnetic signatures and geometric characteristics. This paper presents EarthBridge, a high-fidelity translation framework developed for the 4th Multi-modal Aerial View Image Challenge -- Translation (MAVIC-T). We explore two distinct methodologies: Diffusion Bridge Implicit Models (DBIM), which we generalize using non-Markovian bridge processes for high-quality deterministic sampling, and Contrastive Unpaired Translation (CUT), which utilizes contrastive learning for structural consistency. Our EarthBridge framework employs a channel-concatenated UNet denoiser trained with Karras-weighted bridge scalings and a specialized "booting noise" initialization to handle the inherent ambiguity in cross-modal mappings. We evaluate these methods across all four challenge tasks (SAR $\rightarrow$ EO, SAR $\rightarrow$ RGB, SAR $\rightarrow$ IR, RGB $\rightarrow$ IR), achieving superior spatial detail and spectral accuracy. Our solution achieved a composite score of 0.38, securing the second position on the MAVIC-T leaderboard. Code is available at https://github.com/Bili-Sakura/EarthBridge-Preview.

Figures & tables

Figure 2: Distribution evolution of the diffusion bridge. The source distribution at $\bm{z}_{T}=\bm{x}$ (left) converges through time $t$ to the target distribution at $\bm{z}_{0}=\bm{y}$ (right), governed by the DBIM update rule and denoiser $D_{\theta}$ .

Figure 3: Deterministic DBIM sampling. The noisy input $\bm{z}_{T}=\bm{x}+\bm{\epsilon}_{\text{boot}}$ is progressively denoised by $D_{\theta}(\bm{z}_{t},t,\bm{x})$ to produce the clean output $\bm{z}_{0}=\bm{y}$ . The step sequence shows conditioning, booting noise injection, and gradual refinement from $t=T$ to $t=0$ .

Figure 4: UNet denoiser architecture for 1024px tasks (SAR $\rightarrow$ RGB, RGB $\rightarrow$ IR). The encoder downsamples via ResBlocks and average pooling; the decoder upsamples with skip connections. Source $\bm{x}$ is channel-concatenated with $\bm{z}_{t}$ ; the output is $\hat{\bm{z}}_{0}$ . We train at 512 $\times$ 512 and infer at full resolution, following precedent that diffusion models trained at 512px can still synthesize faithfully at 1024px at inference time [ 37 ] , which greatly reduces training cost versus training directly at 1024px.

Task	Resol.	Ch.	Samp. (orig.)	Samp. (aug.)
SAR $\rightarrow$ EO	256 2	1 $\rightarrow$ 1	89,411	–
SAR $\rightarrow$ RGB	1024 2	1 $\rightarrow$ 3	10,576	75,703
SAR $\rightarrow$ IR	1024 2	1 $\rightarrow$ 1	10,576	74,317
RGB $\rightarrow$ IR	1024 2	3 $\rightarrow$ 1	2,272	7,388

Table 1: Task specifications and training data for the MAVIC-T challenge. Crop augmentation uses 1024

\times

1024 sliding windows with stride 512 (50% overlap); SAR

\rightarrow

optical pairs are georeference-aligned.

Track	Method	FID $\downarrow$	LPIPS $\downarrow$	L1 $\downarrow$	Score $\downarrow$	NFE	Time
SAR $\to$ EO	DBIM	0.22	0.50	0.08	0.27	500	0.42 s
SAR $\to$ RGB	DBIM	0.88	0.64	0.21	0.58	1000	160 s
SAR $\to$ IR	CUT	0.65	0.60	0.15	0.46	1	0.47 s
RGB $\to$ IR	DBIM	0.36	0.15	0.09	0.20	5	1.5 s

Table 2: Main results on the MAVIC-T evaluation set.

\downarrow

lower is better. Time is per-sample inference latency; SAR

\to

EO is run at 256px with large batch size, and its latency is averaged.

Rank	Participant	SAR $\to$ EO	SAR $\to$ RGB	SAR $\to$ IR	RGB $\to$ IR	Combined
1st	HNU-VPAI	0.11	0.50	0.49	0.20	0.32
2nd	EarthBridge ( Ours )	0.27	0.58	0.46	0.20	0.38

Table 3: Leaderboard comparison. Per-task scores (SAR

\to

EO, SAR

\to

RGB, SAR

\to

IR, RGB

\to

IR) from the challenge evaluation.

Figure 5: Qualitative results across all four MAVIC-T translation tasks. Each row corresponds to one task (SAR $\rightarrow$ EO, SAR $\rightarrow$ RGB, SAR $\rightarrow$ IR, RGB $\rightarrow$ IR). Columns show: source image, EarthBridge output, and ground-truth target. Our model preserves structural layout from the source while synthesizing faithful target-modality textures across both 256 $\times$ 256 and 1024 $\times$ 1024 resolutions.

Configuration	SAR $\to$ EO	SAR $\to$ RGB	RGB $\to$ IR
Parameters	$\sim$ 20M	$\sim$ 120M	$\sim$ 120M
VP schedule ( $\beta_{d}$ , $\beta_{\min}$ )	2.0, 0.1	2.0, 0.1	2.0, 0.1
Karras $\rho$	7	7	7
Base channels	64	96	128
Channel mult.	(1,2,3,4)	(1,1,2,2,4,4)	(1,1,2,2,4,4)
Attn. resol.	N/A	32, 16, 8	32, 16, 8

Table 4: DBIM per-task architectural and training settings (SAR

\to

EO, SAR

\to

RGB, RGB

\to

IR).

Configuration	SAR $\to$ IR
Params (total)	$\sim$ 225M
Gen.; Disc.	$\sim$ 180M; $\sim$ 45M
Gen. (ResNet blk.)	9
Discriminator	4-layer PatchGAN
Filters (ngf, ndf)	256
GAN objective	LSGAN

Table 5: CUT architectural and training settings for SAR

\to

IR (ResNet-9-block generator, 4-layer PatchGAN discriminator).

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

RGB $\rightarrow$ IR (NFE sweep)
Task	Method	NFE	FID $\downarrow$	LPIPS $\downarrow$	L1 $\downarrow$	Score $\downarrow$
RGB $\rightarrow$ IR	DBIM	1	0.786	0.293	0.093	0.390
RGB $\rightarrow$ IR	DBIM	2	0.342	0.195	0.091	0.209
RGB $\rightarrow$ IR	DBIM	5	0.357	0.151	0.090	0.200
RGB $\rightarrow$ IR	DBIM	10	0.380	0.150	0.090	0.206
RGB $\rightarrow$ IR	DBIM	20	0.601	0.468	0.076	0.382

Appendix

Table S1 : MAVIC-T test-set competition scores (all tasks). Light-blue rows mark the best composite score per task block (full-row fill).

\downarrow

lower is better. DDBM uses a numerical bridge sampler with the listed steps. CUT rows list generator capacity in the Method column and use NFE

{=}1

(single feed-forward pass). Nano Banana Pro is an external image-generation baseline; we omit NFE (—) when the system does not expose an equivalent step count.

Image-to-Image Translation

Image-to-image translation maps images between visual domains or modalities while preserving relevant content, structure, and semantics.

See all 115 papers

Any2Any: Unified Arbitrary Modality Translation for Remote Sensing
Mar 4, 2026Haoyang Chen, Jing Zhang, Hebaixu Wang +7Remote Sensing Image Understanding Image-to-Image Translation
National Engineering Research Center for Multimedia Software, Institute of Artificial Intelligence, School of Computer Science, Wuhan University, and Hubei Key Laboratory of Multimedia and Network Communication Engineering · Zhongguancun Academy, Beijing, China. 100094 · School of Electronic Information, Wuhan University, Wuhan, China +2
Bridging Modalities and Tasks: A Unified Hierarchical ViT for SAR-to-Optical Translation and Semantic Segmentation
Sep 4, 2026Siyuan Liu, Xuze Zhang, Yongshun Wang +3Image-to-Image Translation Semantic Segmentation
School of Automation, Northwestern Polytechnical University, Xi’an 710129, China · School of Cybersecurity, Northwestern Polytechnical University, Xi’an 710129, China
Earth-OneVision: Extending Remote Sensing Multimodal Large Language Models to More Sensor Modalities and Tasks
Jun 9, 2026Miaoxin Cai, Guanqun Wang, Wei Zhang +6Multimodal Understanding Multimodal Grounding
National Key Laboratory of Science and Technology on Space-Born Intelligent Information Processing (SBIIP), Beijing Institute of Technology, Beijing 100081, China · Advanced Research Institute of Multidisciplinary Sciences, Beijing Institute of Technology, Beijing 100081, China · School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China +5

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

Any2Any: Unified Arbitrary Modality Translation for Remote Sensing

Haoyang Chen, Jing Zhang, Hebaixu Wang, Shiqin Wang+6

National Engineering Research Center for Multimedia Software, Institute of Artificial Intelligence, School of Computer Science, Wuhan University, and Hubei Key Laboratory of Multimedia and Network Communication Engineering · Zhongguancun Academy, Beijing, China. 100094 · School of Electronic Information, Wuhan University, Wuhan, China+2

Remote Sensing Image Understanding Image-to-Image Translation Cross-Modal Representation Transfer Multimodal Diffusion Models Cross-Modal Alignment

Open paper AskPDF arXiv

Multi-modal remote sensing imagery provides complementary observations of the same geographic scene, yet such observations are frequently incomplete in practice. Existing cross-modal translation methods treat each modality pair as an independent task, resulting in quadratic complexity and limited generalization to unseen modality combinations. We formulate Any-to-Any translation as inference over a shared latent representation of the scene, where different modalities correspond to partial observations of the same underlying semantics. Based on this formulation, we propose Any2Any, a unified latent diffusion framework that projects heterogeneous inputs into a geometrically aligned latent space. Such structure performs anchored latent regression with a shared backbone, decoupling modality-specific representation learning from semantic mapping. Moreover, lightweight target-specific residual adapters are used to correct systematic latent mismatches without increasing inference complexity. To support learning under sparse but connected supervision, we introduce RST-1M, the first million-scale remote sensing dataset with paired observations across five sensing modalities, providing supervision anchors for any-to-any translation. Experiments across 14 translation tasks show that Any2Any consistently outperforms pairwise translation methods and exhibits strong zero-shot generalization to unseen modality pairs. Code and models are available at https://github.com/MiliLab/Any2Any.

Abstract

Image-to-Image Translation

Explore similar work

Any2Any: Unified Arbitrary Modality Translation for Remote Sensing

Bridging Modalities and Tasks: A Unified Hierarchical ViT for SAR-to-Optical Translation and Semantic Segmentation

Earth-OneVision: Extending Remote Sensing Multimodal Large Language Models to More Sensor Modalities and Tasks