RealLiFe is a new light field optimization method called that utilizes the sparse manifold of Multi-plane Images (MPI) to generate high-quality light fields in real time. Extensive experiments demonstrate that RealLiFe achieves comparable visual quality to offline methods, but is 100 times faster on average and outperforms other online approaches in terms of performance.

We propose an all-frequency, object-based neural precomputed radiance transfer method to restore global illumination effects. The network efficiently produces low-frequency transfer coefficients for objects, enabling low-frequency illumination. Additionally, we use minimal one-bounce ray-tracing to add high-frequency components, implemented through specular importance sampling. Please note that this work is ongoing, and in this presentation, we will showcase preliminary results and compare them with path-tracing outcomes.