The code to obtain 4-views rendered results of the Cape dataset

[glorioushonor]

Good job! I would like to know the code to obtain rendered results and visibility results of the Cape dataset, as I am interested in obtaining 4-views rendered results of the Cape dataset. Thank you.

[YuliangXiu]

Please check dataset.md to know the rendering procedure

[glorioushonor]

Please check dataset.md to know the rendering procedure So, the THuman2.0 dataset and the CAPE dataset share the same rendering pipeline? However, the CAPE dataset does not have tex_file = f'./data/{dataset}/scans/{subject}/material0.jpeg' and fit_file = f'./data/{dataset}/{smpl_type}/{subject}.pkl'.

[YuliangXiu]

Yes, the textured scans of CAPE are not publicly available.

[glorioushonor]

Yes, the textured scans of CAPE are not publicly available. So I cannot render the CAPE dataset according to my own needs? In fact, I am trying to reproduce the work of D_if, but my test results differ from the paper to some extent. So, I looked into the reasons. I found that Xueting Yang rendered the CAPE dataset from four viewpoints, but I can only download the rendering results of three viewpoints. Could this be a reason for my poorer test results? By the way, is there any way to get the 4-view rendering result of CAPE?

[YuliangXiu]

I think this is a typo, D-IF does evaluation under the same setting as ICON with 3 views (0, 120, 240), see PIFuDataset.py#L99.

All these testing data could be downloaded from CAPE. No need to re-render them.

I will ask @yxt7979 to update the arXiv later. Thanks for correcting us.