The study focuses on the development of HDR-HexPlane, a dynamic High Dynamic Range (HDR) Neural Radiance Fields (NeRF) framework. Unlike traditional NeRF methods that use low dynamic range images, HDR-HexPlane utilizes images with various exposures to create high-quality images from dynamic scenes. The framework incorporates a learnable exposure mapping function and a camera response function for stable learning. The researchers also provide a dataset with multiple dynamic scenes captured with different exposures for evaluation. The method allows rendering of high-quality novel-view images at any desired exposure.
Publication date: 11 Jan 2024
Project Page: https://guanjunwu.github.io/HDR-HexPlane/
Paper: https://arxiv.org/pdf/2401.06052
