Feature list:
- Build a deep learning-based SLAM framework, including unified modules for dataset reading, configuration parsing, evaluation, and visualization.
- Implement tracker, mapper, and visualizer using multiprocessing.
- Support downloading multiple datasets, including replica, neural_rgbd_data, tum_rgbd, and apartment.
- Provide online and offline visualization.
- Support evaluation of various metrics for algorithms, including trajectory accuracy (ATE [cm]), rendering image quality (PSNR[dB], SSIM, LPIPS), and reconstruction quality (Precision[%], Recall[%], F_score[%], Depth L1[cm], Accuracy [cm], Completion [cm], Completion ratio[%]).
- Seven algorithms have been integrated into the framework: NICE-SLAM, Co-SLAM, Vox-Fusion, Point-SLAM, SplaTAM, DPVO, NeuralRecon.