Reproducing the results for table 3

[arnobanu]

Dear authors, thank you for sharing your valuable work. I was trying to generate the results like table 3 of the paper using the script GIF/tests/deca_inf_vs_given_cond_landmark_viz.py. Here, I could not find the paths for flame_datapath and face_region_only_indices variable.

[ParthaEth]

I seem to have cleared resources and do not have the file you are looking for anymore. The file contained the index of the vertices of FLAME that make up the front part of the face. The idea was to evaluate the mean error between the 3D mesh corresponding to the given parameters and a mes obtained by running DECA on the generated image. However, since the back of the head and the head under the hair is not observed, it is good to exclude those parts from contributing and possibly dominating the error. You can manually select the frontal part of the face in Blender and dump the vertex indices to get a similar result.