Page 20 of

Each model is trained for 300K iterations. We adopt L = 16 as the number of frequency bands of our model. In Figure 1, an ablation experiment for adaptive transparency is shown. We selected the Lego model from the NeRF-synthetic dataset, the skull model (scan65) from the DTU dataset and the dog model from the BlendedMVS for validation. In this section, more qualitative results are provided.