AdamSLAM is a project that deals with generating a 3D model of a real scene based on photos of the scene.
3D Gaussian Splatting (3DGS) is a major breakthrough in this field. That algorithm inputs are the photos of the scene, as well as camera parameters (camera focal length and position for each photo). The process of identifying these parameters through Structure-from-Motion (SfM) is called calibration.
The quality of the camera calibration is of major importance for evaluating progresses in novel view synthesis, as a 1 pixel error on the calibration has a significant impact on the reconstruction quality. While there is no ground truth for real scenes, the quality of the calibration is assessed by the quality of the novel view synthesis. This project proposes to use a 3DGS model to fine tune calibration by backpropagation of novel view color loss with respect to the cameras parameters. The new calibration alone brings an average improvement of 0,4 dB PSNR on the dataset used as reference by 3DGS. The fine tuning may be long and its suitability depends on the criticity of training time, but for calibration of reference scenes, such as, the stake of novel view quality is the most important.
The project is based on a fork and is available on GitHub.