SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM

Rendering: 400 FPS Figure 1. SplaTAM enables precise camera tracking and high-fidelity reconstruction for dense simultaneous localization and mapping (SLAM) in challenging real-world scenarios. SplaTAM achieves this by online optimization of an explicit volumetric representation, 3D Gaussian Splatting [14], using differentiable rendering. Left: We showcase the high-fidelity 3D Gaussian Map along with the train (SLAM-input) & novel view camera frustums. It can be noticed that SplaTAM achieves sub-cm localization despite the large motion between subsequent cameras in the texture-less environment. This is particularly challenging for state-of-the-art baselines leading to the failure of tracking. Right: SplaTAM enables photo-realistic rendering of both train & novel views at 400 FPS for a resolution of 876 × 584.