STAR++: A Physical-Aware Benchmark and Framework for Astronomical Super-Resolution across Multiple Filters
Authors: Guohang Zhuang, Jiaze Li, Kuo-cheng Wu, Hao Du, Jinyang Huang, Xiang Zhang, Wanli Ouyang, Yan Lu
Organizations: School of Computer Science and Information Engineering, Hefei University of Technology, Hefei 230026, China · Shanghai Artificial Intelligence Laboratory, Shanghai 200240, China · University of Science and Technology of China, Hefei · The Chinese University of Hong Kong, Hong Kong · School of Cyber Security, Tianjin University, Tianjin, China
Astronomical super-resolution (ASR) aims to reconstruct high-resolution star fields from low-resolution telescope observations across multiple filters for astrophysical analysis. However, developing a physics-aware ASR framework faces two challenges: 1) Data view: no large-scale benchmark supports field-level ASR, especially on unseen filters. 2) Method view: Existing ASR methods often fail to preserve stellar flux across spatial resolutions. Moreover, ASR requires reconstructing high-frequency components, including accurate star locations and profiles, making it more ill-posed than natural-image SR. To tackle these problems, we introduce STAR++, a large-scale benchmark comprising approximately 1.7 million flux-consistent image pairs across 36 filters, derived from Hubble Space Telescope (HST) observations. STAR++ uses a flux-preserving data generation pipeline to ensure photometric consistency across resolutions and provides the Flux Error (FE) metric to evaluate ASR models from a physical perspective. To address method-level challenges, we propose a Flux-Invariant Super-Resolution Plus (FISR++) model that generalizes to unseen filters while accurately reconstructing flux-consistent high-resolution images. Beyond the flux-preserving modules, FISR++ introduces two novel components. Specifically, a Filter-Aware Adapter (FAA) incorporates filter metadata to adapt image representations to varying spectral and optical conditions. In addition, a Spatial Ambiguity Disentangle Loss (SADL) models spatial ambiguity as pixel-shift uncertainty, mitigating the smoothing of learned conditional mean reconstructions caused by this ambiguity. Extensive experiments demonstrate that FISR++ consistently outperforms existing ASR methods in reconstruction quality and flux accuracy, while generalizing to unseen filters, highlighting its effectiveness for photometrically faithful ASR.
Figures & tables
Fig. 1: Comparison of previous datasets and ours, highlighting the richer astronomical structures and phenomena captured by STAR++, such as cross-object interactions, weak lensing, Dense star fields, and spiral galaxies. The STAR++ examples are visualized as RGB composites, with red , green , and blue filter labels indicating the filters mapped to the corresponding color channels.
Sky coordinates: RA 0.07∘ ~ 359.94∘ ; Dec −89.13∘ ~ 87.41∘
Split
Filter(s)
Pivot. λ (nm)
Frac. BW
Images
LR–HR pairs
Training (seen)
29 filters
270.3–843.7
0.004–0.352
1,008
922,586
Validation (seen)
29 filters
270.3–843.7
0.004–0.352
237
162,593
Test (unseen-interp.)
F435W
432.9
0.069
82
76,019
F555W
533.6
0.082
93
47,169
TABLE II: Dataset statistics, data splits, and filter characteristics of STAR++, including pivot wavelength and fractional bandwidth. Unseen filters are divided into interpolation and extrapolation subsets according to their pivot wavelengths relative to the training wavelength range.
Fig. 2: Schematic diagram of the flux-consistent downsampling process. The workflow illustrates the transformation of HR and LR image pixels into the celestial coordinate system, the computation of overlapping sky regions between HR and LR receptive fields, and the flux calculation for LR pixels using weighted contributions from overlapping HR pixels.
Fig. 3: Overview of the proposed FISR++ framework. The LR input is first transformed into the asinh intensity space and processed by an encoder and a Flux Guidance Generation (FGG) module. The extracted flux guidance is incorporated into multi-scale features through Flux Guidance Controllers (FGCs), while the Filter-Aware Adapter (FAA) integrates physical observation priors to enhance feature representations. The enhanced features are progressively decoded and upsampled to reconstruct the HR image. During training, a pixel-wise uncertainty map is also predicted for the Spatial Ambiguity Disentangle Loss (SADL).
Fig. 4: Illustration of the Spatial Ambiguity Disentanglement Loss (SADL). During training, SADL learns a global shift δ under a Gaussian prior and applies a first-order Taylor correction to the prediction. The curves illustrate pixel-value profiles, where the vertical distances between the prediction and ground truth represent pixel-wise L1 errors before and after shift correction, weighted by the predicted uncertainty.
Training set
Method
PSNR (dB) ↑
SSIM ↑
FE ↓
Interp.
Extrap.
Overall
–
Bicubic
29.72
23.53
27.06
0.7585
2.8908
STAR
EDSR
31.94
24.16
28.61
0.7934
1.2957
RCAN
33.86
26.79
30.83
0.8201
0.7404
SwinIR
34.79
26.95
31.43
0.8264
0.7141
HAT
35.14
26.76
31.55
0.8253
0.6205
TABLE III: ASR performance on seven unseen STAR++ filters at ×2 . The interpolation subset comprises F435W, F555W, F606W, and F625W, while the extrapolation subset comprises F218W, F225W, and F850LP. STAR and STAR++ indicate the training datasets.
Training set
Method
PSNR (dB) ↑
SSIM ↑
FE ↓
Interp.
Extrap.
Overall
–
Bicubic
26.23
20.41
23.74
0.6523
4.4267
STAR
EDSR
30.84
23.44
27.67
0.7165
1.0674
RCAN
31.53
23.67
28.16
0.7234
0.8701
SwinIR
31.68
24.03
28.40
0.7283
0.8273
HAT
31.99
24.03
28.58
0.7328
0.7659
TABLE IV: ASR performance on unseen STAR++ filters at ×4 . STAR and STAR++ indicate the training datasets.
Dataset
ID
FISR
FISR++
Performance
FGG
FGC
FCL
FAA
SADL
Asinh
PSNR ↑
SSIM ↑
FE ↓
UL
SSR
STAR
(a)
-
-
-
-
-
-
-
37.77
0.8288
0.7022
(b)
-
-
✓
-
-
-
-
37.86
0.8283
0.6389
(c)
✓
-
-
-
-
-
-
37.71
0.8286
0.7572
(d)
✓
-
✓
-
-
-
-
37.87
0.8301
0.7467
TABLE V: Ablation studies of different components under the ×2 super-resolution setting. UL and SSR denote Uncertainty Loss and Spatial Shift Regularization, respectively.
σ
PSNR (dB) ↑
SSIM ↑
FE ↓
0.05
35.37
0.8618
0.4378
0.10
35.39
0.8619
0.4347
0.15
35.42
0.8623
0.4341
0.20
35.41
0.8622
0.4343
0.50
34.94
0.8607
0.4401
TABLE VI: Sensitivity analysis of the Gaussian prior standard deviation σ in SADL. The best results are highlighted in bold.
Scale
Flux Loss
STAR
STAR++
EDSR [ 8 ]
RCAN [ 24 ]
SwinIR [ 28 ]
HAT [ 29 ]
FISR
FISR++
FISR++
×2
PSNR ↑ ( w/o )
35.74
37.46
37.60
37.91
38.01
38.09
35.28
SSIM ↑ ( w/o )
0.8086
0.8277
0.8279
0.8296
0.8313
0.8346
0.8605
FE ↓ ( w/o )
1.3249
0.7451
0.7593
0.6918
0.5775
0.5469
0.4410
PSNR ↑ ( w/ )
35.89
37.89
37.13
37.94
38.27
38.83
35.42
SSIM ↑ ( w/ )
0.8092
0.8286
0.8281
0.8303
0.8334
0.8384
0.8623
TABLE VII: Comparison of different methods with and without FCL under ×2 and ×4 ASR settings on STAR and STAR++. Bold entries indicate that the setting with FCL (w/) improves over the setting without FCL (w/o) for the same metric and method.
Fig. 5: Visualization of SR reconstruction results on unseen interpolation filters. For each filter, the upper row shows the reconstructions from different methods, while the lower row presents the corresponding residual maps.
Fig. 6: Visualization of SR reconstruction results on unseen extrapolation filters. For each filter, the upper row shows the reconstructions from different methods, while the lower row presents the corresponding residual maps.
Fig. 7: Visualization of stellar centroid shifts under different SR methods. Left: six selected sources marked by boxes. Right: comparison of their centroid offsets across different SR methods.