Trini: An Efficient Representation of Dynamic Scenes for Sparse-View Camera Settings

3D reconstruction of a dynamic scene has been a challenging task in vision. Several strategies have been developed to enhance the reconstruction of dynamic scenes, with some employing tri-projection decomposition techniques that surpass D-NeRF in terms of speed and effectiveness. This paper introduces Trini, which decomposes a dynamic 3D scene into three volumes dealing with the 3D coordinates influenced by time. Each volume is further structured with four marginalized planes. These planes are then integrated with a compact MLP for rendering superior results in a seamless manner. Additionally, we incorporate a technique to efficiently determine coordinates in a set of distinct images for enhancing the reconstruction process for cases involving sparse-view camera images. The efficacy of our method outperforms other state-of-the-art techniques and is particularly evident in capturing the dynamic elements and edges present in the scene.