Supplementary Material A Comprehensive Benchmark for Neural Human Radiance Fields

Due to space constraints in the main paper, We elaborate on the following in the supplementary material: (1) more ablation study of different settings in Sec. For multi-view video sequences, consecutive frames may have severe information redundancy that is adverse to the reconstruction of the human geometry and appearance. Thus, we fix the number of train frames and train views but set the frame sampling interval to be 1, 5 and 10. The experimental results are shown in Tab. 1. In general, with larger sampling interval, the performance of selected representative scene-specific methods would increase.