3D Reconstruction of Plant Based on NeRF

At present, aiming at the problems of quality and lack of modeling methods in plant 3D modeling technology, an improved fast 3D plant modeling technology based on Neural radiation field (NeRF) is proposed. Taking the simple plant multi view image as the input, the pose of the multi view image is obtained by using SFM. The plant image and pose information are sent to the improved nerf model for training, and the high-quality plant 3D scene is obtained. In order to improve the speed of 3D modeling, this study proposes an improved method of nerf neural radiation field based on instant NGP to reconstruct 3D plants, and a small fusion MLP based on spherical harmonics and multi-resolution hash coding for training and rendering. Compared with the original nerf training time, this method also achieves good rendering quality.