Mono4DGS-HDR: High Dynamic Range 4D Gaussian Splatting from Alternating-exposure Monocular Videos
Jinfeng Liu, Lingtong Kong, Mi Zhou, Jinwen Chen, Dan Xu
2025-10-22
Summary
This paper introduces a new system called Mono4DGS-HDR that creates realistic, 4D high dynamic range (HDR) scenes from regular videos. Think of it like taking a normal video and turning it into something that looks incredibly vibrant and lifelike, with details you wouldn't normally see.
What's the problem?
Creating a 4D HDR scene from a standard video is really hard. Normal videos don't capture the full range of light and color that exists in the real world, and usually, you need multiple cameras or special equipment to do so. This research tackles the challenge of building a detailed HDR scene from just *one* regular video, without knowing the camera's exact movements during filming. Existing methods struggle with both the quality of the resulting scene and how quickly it can be created.
What's the solution?
The researchers developed a two-step process using a technique called Gaussian Splatting. First, they build a basic HDR representation of the video without needing to know where the camera was pointing. Then, they refine this representation, placing everything in a 3D world space and figuring out the camera's movements at the same time. They also added a clever trick to make sure the brightness stays consistent throughout the video, avoiding flickering or sudden changes. Essentially, they're building a 3D model of the scene from the video, but one that accurately represents light and color.
Why it matters?
This work is important because it opens the door to creating high-quality HDR scenes from everyday videos. This could have big implications for things like virtual reality, special effects in movies, and even just improving the way we view and share videos online. Because it's faster and requires less equipment than previous methods, it makes HDR scene creation more accessible.
Abstract
We introduce Mono4DGS-HDR, the first system for reconstructing renderable 4D high dynamic range (HDR) scenes from unposed monocular low dynamic range (LDR) videos captured with alternating exposures. To tackle such a challenging problem, we present a unified framework with two-stage optimization approach based on Gaussian Splatting. The first stage learns a video HDR Gaussian representation in orthographic camera coordinate space, eliminating the need for camera poses and enabling robust initial HDR video reconstruction. The second stage transforms video Gaussians into world space and jointly refines the world Gaussians with camera poses. Furthermore, we propose a temporal luminance regularization strategy to enhance the temporal consistency of the HDR appearance. Since our task has not been studied before, we construct a new evaluation benchmark using publicly available datasets for HDR video reconstruction. Extensive experiments demonstrate that Mono4DGS-HDR significantly outperforms alternative solutions adapted from state-of-the-art methods in both rendering quality and speed.