SAD-SLAM

For the final project of the class of 3D Computer Vision with Deep Learning Applications, our team proposed “SAD-SLAM: Sign-Agnostic Dynamic Simultaneous Localization and Mapping”. This project aims to optimize mapping and tracking, and remove dynamic objects. It improves the performance of mapping and tracking in NICE SLAM using Sign-Agnostic optimization. Dynamic objects removal is implemented in NICE SLAM using Mask R-CNN and background inpainting. The results are validated on some datasets and experiments. More details could be found on Github.