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

TL;DR

This work tackles photometric calibration of the high-resolution JunoCam by translating its images to the HST photometric domain using unpaired data. It introduces SP-I2I, a Schrödinger Bridge-based I2I framework that decouples spectral color transfer from spatial detail preservation via a learnable Laplacian gate and multi-scale Laplacian losses, enabling preservation of meso-scale Jovian features. A new JunoCam2HST dataset is presented to train and evaluate the method, demonstrating superior spatial fidelity and competitive spectral alignment compared with state-of-the-art unpaired translation baselines. The approach also shows promise for unpaired pansharpening in remote sensing, highlighting broader applicability to heterogeneous data fusion in planetary science.

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, wide coverage during perijove passes, and a long baseline. However, JunoCam lacks absolute photometric calibration, hindering quantitative analysis of the Jovian atmosphere. 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, focusing on addressing the resolution discrepancy between the two sensors. 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, importantly, we show the broader impact of our proposed solution on relevant remote sensing data for the pansharpening task.

Figures (10)

• Figure 1: A comparison of naive vs. structure-preserving translation for calibrating high-resolution JunoCam images with low-resolution HST data. While Naive I2I methods lose fine storm details, our Structure-Preserving I2I method successfully retains this critical, high-frequency information.
• Figure 2: Our network disentangles the learning process into two streams. First, a gating network acts as a dynamic low-pass filter, reweighing the predicted image to learn low-frequency color from the HST domain. Simultaneously, a Spatial Consistency Loss explicitly retains the high-frequency spatial detail from the original JunoCam source.
• Figure 3: Qualitative comparison of our method (SP-I2I) against baselines. The JunoCam (Source) images exhibit rich, high-frequency detail but are uncalibrated, while the HST (Target) images show the desired calibrated color but lack this detail. Note that these images are spatially and temporally unpaired but are sampled from the same latitudinal zones. Baseline methods (CUT, UNSB) adopt the target color but fail to preserve these 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.
• Figure 4: Average Power Spectral Density (PSD) comparison. While the baseline UNSB (green) mimics the low-detail HST target (orange), our method, SP-I2I (red), uniquely matches the HST low-frequency profile while also preserving the high-frequency detail from the JunoCam source (blue). Note that $f_s/2$ is the Nyquist frequency for images.
• Figure 5: Qualitative comparison of our method (SP-I2I) against unpaired baselines on the QuickBird dataset, assuming no co-registration.