Page 20 of

It can be seen that when the parameter is too small, the surface reconstruction tends to be oversmoothed. First row: the reference image and the reconstructed surface using different coarse-to-fine parametersαd withL = 16. We selected the Lego model from the NeRF-synthetic dataset, the skull model (scan65) from the DTU dataset and the dog model from theBlendedMVS forvalidation. First row: the reconstructed surface without the adaptivetransparencystrategy. "OURBase-Sigmoid"shows better geometry consistencyon the roof ofthe house compared withVolSDF and NeuS.