Expo-GS: Exposure-Aware Signed Distance Function in Gaussian Splatting for High Dynamic Range

High dynamic range novel view synthesis (HDR-NVS) remains challenged by geometric artifacts and radiometric distortions under multi-exposure conditions, primarily due to existing methods ignoring exposure and over-relying on color cues. Inspired by the integrated processing of color and structure of the human visual system (HVS), we propose Expo-GS, a novel framework that decomposes HDR-NVS into three interpretable components, namely, Irradiance Field Training, Geometry Field Training, and Interactive Joint Training. Central to Expo-GS is the exposure-aware signed distance function (Expo-SDF), which dynamically reweights geometric supervision via localized exposure reliability estimation, suppressing noisy gradients from unstable regions while enhancing structure learning in well-exposed areas. Building on this, we design an interactive optimization strategy that synchronizes Gaussian primitive growth and pruning with evolving Expo-SDF cues, enabling exposure-aware density control and eliminating hallucinated structures near exposure transitions. Experiments show that Expo-GS significantly outperforms prior methods on both synthetic and real-world datasets. It achieves a peak PSNR of 39.06 dB under HDR settings and up to 41.38 dB in the LDR-OE configuration, excelling in preserving high-frequency textures and maintaining structural consistency.

Creating realistic 3D scenes from a few ordinary photos lets a computer show a place from viewpoints the camera never captured — the technology behind virtual reality, self-driving cars, and digital recreations of real spaces. This gets hard when a scene has very bright and very dark areas at once, like a sunlit window beside a shadowy corner. Ordinary photos can't record that full range of brightness, so existing methods often invent fake or blurry shapes in regions that are washed out or too dark. Our method, Expo-GS, takes inspiration from human vision, which handles color and shape somewhat separately and trusts most the parts of a scene it can see clearly. Following this idea, our system learns a scene's colors and its 3D structure separately, then combines them — judging how trustworthy each region of every photo is from how well it is exposed and building structure from well-lit areas while ignoring misleading over- or under-exposed ones. The result is sharper, more faithful 3D scenes under difficult lighting than previous approaches, while staying fast and efficient.