Parallel Inversion of Neural Radiance Fields for Robust Pose Estimation

We present a parallelized optimization method based on fast Neural Radiance Fields (NeRF) for estimating 6-DoF target poses. Given a single observed RGB image of the target, we can predict the translation and rotation of the camera by minimizing the residual between pixels rendered from a fastNeRF model and pixels in the observed image. We integrate a momentum-based camera extrinsic optimization procedure intoInstant Neural Graphics Primitives, a recent exceptionally fastNeRF implementation. By introducing parallel Monte Carlo sampling into the pose estimation task, our method over comes local minima and improves efficiency in a more extensive search space. We also show the importance of adopting a more robust pixel-based loss function to reduce error. Experiments demonstrate that our method can achieve improved generalization and robustness on both synthetic and real-world benchmarks.