AS-Bridge: A Bidirectional Generative Framework Bridging Next-Generation Astronomical Surveys

Dichang Zhang; Yixuan Shao; Simon Birrer; Dimitris Samaras

astro-ph.IMMar 12, 2026

AS-Bridge: A Bidirectional Generative Framework Bridging Next-Generation Astronomical Surveys

Authors: Dichang Zhang, Yixuan Shao, Simon Birrer, Dimitris Samaras

Organizations: Stony Brook University, Stony Brook, NY, USA

Abstract

The upcoming decade of observational cosmology will be shaped by large sky surveys, such as the ground-based LSST at the Vera C. Rubin Observatory and the space-based Euclid mission. While they promise an unprecedented view of the Universe across depth, resolution, and wavelength, their differences in observational modality, sky coverage, point-spread function, and scanning cadence make joint analysis beneficial, but also challenging. To facilitate joint analysis, we introduce A(stronomical)S(urvey)-Bridge, a bidirectional generative model that translates between ground- and space-based observations. AS-Bridge learns a diffusion model that employs a stochastic Brownian Bridge process between the LSST and Euclid observations. The two surveys have overlapping sky regions, where we can explicitly model the conditional probabilistic distribution between them. We show that this formulation enables new scientific capabilities beyond single-survey analysis, including faithful probabilistic predictions of missing survey observations and inter-survey detection of rare events. These results establish the feasibility of inter-survey generative modeling. AS-Bridge is therefore well-positioned to serve as a complementary component of future LSST-Euclid joint data pipelines, enhancing the scientific return once data from both surveys become available. Data and code are available at https://github.com/ZHANG7DC/AS-Bridge.

Figures & tables

Figure 1. Overview of AS-Bridge. The central panel shows the visible sky, with the LSST and Euclid survey footprints marked in blue and red, respectively. The overlapping regions indicate areas jointly observed by the ground-based LSST (left) and the space-based Euclid mission (right). LSST provides multi-band optical images that are more blended due to atmospheric seeing, while Euclid delivers sharper near-infrared observations from space. From these overlapping regions, matched image cutouts are extracted and used to train AS-Bridge, which models the probabilistic translation between the two survey domains using a Brownian bridge formulation. Diagram showing LSST and Euclid survey footprints, paired image cutouts, and the AS-Bridge Brownian bridge translation process between the two survey domains.

Figure 2. LSST-to-Euclid translation. Left: LSST where nearby galaxies are strongly blended due to atmospheric seeing and PSF broadening. Middle: Euclid observation revealing the true multi-object structure at higher spatial resolution. Right: AS-Bridge reconstruction from the LSST input, correctly recovering the number of galaxies and accurately localizing their centers. Three-panel comparison of a blended LSST galaxy cutout, the corresponding sharper Euclid observation, and the AS-Bridge Euclid reconstruction.

Figure 3. Euclid-to-LSST translation with multiple realizations. From left to right: LSST observation, Euclid observation, and four stochastic LSST reconstructions generated by AS-Bridge from the Euclid input. While Euclid provides high-resolution morphology but limited spectral information, AS-Bridge reconstructs plausible LSST multi-band appearances that preserve the underlying structure and produce color variations consistent with the ground truth. Row of image cutouts comparing a ground-truth LSST observation, the Euclid input, and multiple stochastic LSST reconstructions generated by AS-Bridge.

Figure 4. Midpoint-to-Euclid AS-Bridge reconstructions for (a) a system containing a common in-distribution edge-on disk galaxy and (b) a rare galaxy–galaxy strong-lensing system. From left to right: LSST , Euclid , and AS-Bridge reconstruction of Euclid from midpoint. Although the two cases appear visually similar, AS-Bridge faithfully reconstructs the elongated disk morphology in (a), while in (b) it fails to reproduce the extended arc. This facilitates rare event detection. Two qualitative examples comparing LSST input, Euclid observation, and midpoint-to-Euclid AS-Bridge reconstructions for a regular edge-on galaxy and a rare strong-lensing system.

Non-diffusion baselines
Method	LSST $\to$ Euclid					Euclid $\to$ LSST
Method	CRPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	MAE $\downarrow$	RMSE $\downarrow$	CRPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	MAE $\downarrow$	RMSE $\downarrow$
SPADE ( Park et al., 2019 )	3.39	16.77	0.56	0.110	0.146	16.52	11.09	0.06	0.260	0.392
OASIS ( Schonfeld et al., 2021 )	4.65	16.78	0.57	0.110	0.146	13.33	11.34	0.10	0.250	0.352
pix2pix ( Isola et al., 2017 )	4.35	16.95	0.60	0.110	0.143	73.03	12.81	0.07	0.210	0.289
Diffusion baselines

Table 1. Bidirectional translation metrics.

Test case	Method	FPR@1% TPR $\downarrow$	FPR@5% TPR $\downarrow$	AUPR $\uparrow$
Strong lensing	AS-Bridge	0.00%	0.18%	0.80
	Deco-Diff ( Beizaee et al., 2025 )	1.1%	5.0%	0.61
	CFM ( Costanzino et al., 2024 )	0.24%	1.2%	0.75
HST anomaly	AS-Bridge	0.00%	0.00%	1.000
	Deco-Diff ( Beizaee et al., 2025 )	45.71%	63.13%	0.3297
	CFM ( Costanzino et al., 2024 )	0.00%	0.00%	0.962

Table 2. Rare-event detection across two anomaly test cases.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Train samples	CRPS E $\to$ L	CRPS L $\to$ E
$1{,}000$	34.10	7.79
$10{,}000$	11.60	5.31
$20{,}000$	8.65	2.95
$40{,}000$	7.96	2.70
$60{,}000$	7.93	2.50
$80{,}000$	7.92	2.43

Appendix

Table 3. Ablation on training-set size.

Steps	CRPS E $\to$ L	CRPS L $\to$ E	Samples/s
20	9.68	2.69	40.87
50	7.90	2.38	16.33
100	7.88	2.30	8.13
200	7.90	2.31	4.06

Appendix

Table 4. Ablation on reverse sampling steps.

Conditional Generative Modeling

Conditional generative modeling learns to produce data distributions steered by labels, inputs, text, or physical and structural constraints.

See all 380 papers

Amplifying the imaging power of digital sky surveys with space telescopes data and generative AI
Aug 21, 2026Sai Teja Erukude, Lior ShamirImage Enhancement Image Super-Resolution
Kansas State University, Manhattan, KS, 66506, USA
COP-GEN: Latent Diffusion Transformer for Copernicus Earth Observation Data
Mar 3, 2026Miguel Espinosa, Eva Gmelich Meijling, Valerio Marsocci +2Conditional Generative Modeling Multimodal Diffusion Models
School of Engineering, University of Edinburgh · European Space Agency (ESA) · Asterisk Labs
MetaEarth-MM: Unified Multimodal Remote Sensing Image Generation with Scene-centered Joint Modeling
May 19, 2026Zhiping Yu, Chenyang Liu, Jinqi Cao +4Cross-Modal Learning Unified Multimodal Models
Department of Aerospace Intelligent Science and Tech-nology, School of Astronautics, Beihang University, Beijing 100191, China · State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing 100191, China · Shenyuan Honors College, Beihang University, Beijing 100191, China

Aug 21, 2026J/K move · Enter open · S save

Amplifying the imaging power of digital sky surveys with space telescopes data and generative AI

Sai Teja Erukude, Lior Shamir

Kansas State University, Manhattan, KS, 66506, USA

Image Enhancement Image Super-Resolution Generative Modeling

Open paper AskPDF arXiv

While Digital sky surveys provide excellent throughput of image data and can cover a large footprint, their imaging power is normally inferior to that of space-based telescopes. Space-based telescopes, on the other hand, provide excellent imaging power and can image the deep Universe, but cannot provide the same throughput as advanced ground-based sky surveys. Here, we utilize generative AI to elevate the quality of galaxy images taken by ground-based telescopes to the level of details enabled by space telescopes. The solution is based on the nature of galaxy shapes, allowing generative AI trained on space-based images to convert weak signal into detailed and clear galaxy images. The method allows for combining the high throughput of ground-based sky surveys with the image quality of space-based telescopes. The source code for the method is available, as well as paired training data and a catalog of 63,202 galaxy images enhanced by the proposed method. We also provide a software tool that encapsulates the entire pipeline and the custom generative AI model to generate galaxy images with enhanced quality.

Abstract

Conditional Generative Modeling

Explore similar work

Amplifying the imaging power of digital sky surveys with space telescopes data and generative AI

COP-GEN: Latent Diffusion Transformer for Copernicus Earth Observation Data

MetaEarth-MM: Unified Multimodal Remote Sensing Image Generation with Scene-centered Joint Modeling