DyLiN: Making Light Field Networks Dynamic

Render time: 3 s TiNeuVox [8] Render time: 7 s Ours Render time: 0.1 s Figure 1. Our proposed DyLiN for dynamic 3D scene rendering achieves higher quality than its HyperNeRF teacher model and the stateof-the-art TiNeuVox model, while being an order of magnitude faster. Right: DyLiN is of moderate storage size (shown by dot radii). For each method, the relative improvement in Peak Signal-to-Noise Ratio over NeRF (∆PSNR) is measured for the best-performing scene.