Structure-Preserving Unpaired Image Translation to Photometrically Calibrate JunoCam with Hubble Data

Abstract

Insights into Jupiter's atmospheric dynamics are vital for understanding planetary meteorology and exoplanetary gas giant atmospheres. To study these dynamics, we require high-resolution, photometrically calibrated observations. Over the last 9 years, the Juno spacecraft's optical camera, JunoCam, has generated a unique dataset with high spatial resolution, but it lacks absolute photometric calibration, hindering quantitative analysis.

Using observations from the Hubble Space Telescope (HST) as a proxy for a calibrated sensor, we present a novel method for performing unpaired image-to-image translation (I2I) between JunoCam and HST. Our structure-preserving I2I method, SP-I2I, incorporates explicit frequency-space constraints designed to preserve high-frequency features ensuring the retention of fine, small-scale spatial structures—essential for studying Jupiter's atmosphere.

We demonstrate that state-of-the-art unpaired image-to-image translation methods are inadequate to address this problem, and we show the broader impact of our proposed solution on relevant remote sensing data for the pansharpening task.

Method: Structure-Preserving I2I

Our primary challenge is to apply the scientifically calibrated color profile of HST observations to high-resolution JunoCam images without sacrificing JunoCam's superior spatial detail. Standard adversarial training tends to penalize fine-scale JunoCam structures as out-of-domain artifacts relative to the lower-resolution HST target.

Our network (Figure above) disentangles the learning process into two streams:

Gating Network: Acts as a dynamic low-pass filter, reweighing the predicted image to learn low-frequency color from the HST domain.
Spatial Consistency Loss: Explicitly retains the high-frequency spatial detail from the original JunoCam source via a Laplacian pyramid loss.

Quantitative Results

We evaluate our method against state-of-the-art unpaired translation methods (UNSB and CUT). SP-I2I achieves superior spatial fidelity (highest SSIM and PSNR) while maintaining a competitive spectral alignment with the HST target.

Method	χ² ↓	SSIM ↑	PSNR ↑
UNSB	0.0704	0.7450	19.7700
CUT	0.0430	0.7092	19.2751
SP-I2I (Ours, L=6)	0.0681	0.9453	26.5560

Metrics on JunoCam2HST test set. Spatial metrics (SSIM, PSNR) are computed against the source JunoCam image, Spectral metrics (&chi²) are computed again target HST images.

Qualitative Results

Comparison with Baselines

The baseline methods (CUT, UNSB) adopt the target color but fail to preserve fine spatial details, resulting in blurry outputs. In contrast, SP-I2I successfully retains the sharp source structures while fusing them with the calibrated HST color profile.

Comparisons against baselines. SP-I2I retains the high-frequency cloud bands visible in the Source (JunoCam) while matching the Target (HST) color profile.

Mosaic Reconstruction (Perijove 18)

We visually compare the calibration performance on the held-out Perijove 18 dataset. The baselines blur the rich atmospheric structures, whereas our method preserves them. Click on the images below to view full resolution.

Original JunoCam
(Uncalibrated, High Detail)

UNSB Baseline
(Blurry Details)

Ours (SP-I2I)
(Calibrated + High Detail)

BibTeX


    @misc{singh2025structurepreservingunpairedimagetranslation,
      title={Structure-Preserving Unpaired Image Translation to Photometrically Calibrate JunoCam with Hubble Data},
      author={Aditya Pratap Singh and Shrey Shah and Ramanakumar Sankar and Emma Dahl and Gerald Eichstädt and Georgios Georgakis and Bernadette Bucher},
      year={2025},
      eprint={2511.22668},
      archivePrefix={arXiv},
      primaryClass={astro-ph.IM},
      url={https://arxiv.org/abs/2511.22668},
}