3D Mesh Model Generation from 2D Images for Small Furniture Items

The proposed 3D reconstruction system is a novel approach that takes a video input and outputs a high-quality 3D mesh model that can be rendered and viewed through augmented reality. The system's novelty lies in its ability to obtain a high-quality optimized mesh using a simple app, providing a balance between speed and accuracy. This makes the system suitable for various potential applications, ranging from assisting buyers and sellers with their shopping, whether online or in physical stores, to being a tool for scanning and acquiring 3D models for use in 3D printing. The research details how best to use Nvidia's Instant NeRF (neural radiance field) model and how to prepare the data for training using multiple pre-processing steps to isolate the object of interest and limit the scene for faster training with less computing power. Post-processing steps are also employed to further enhance the model. A dataset was created using the system, which had structural similarity index (SSIM) measure values of at least over 0.5 within a random sample size.