Review for NeurIPS paper: Geo-PIFu: Geometry and Pixel Aligned Implicit Functions for Single-view Human Reconstruction

Weaknesses: 1. Evaluation setup is not completely clear. Authors do not mention whether they re-trained the baselines (PIFu and DeepHuman) with their images or directly used the pre-trained models. This information is critical as the training data (number of samples and quality of scans) for PIFu was very different than the DeepHuman dataset. In line 290, authors write that they add several FC layers to the fusion network. It is not clear whether the number of parameters were kept the same for Exp-a,b.