About cycle consistency loss

[fnzhan]

Hi @fbxiang , thanks for sharing the codes. I did some simple experiments, and it seems that removing the cycle loss leads to better novel view synthesis (but maybe more distorted uv map). In Fig. 3 of the paper, removing cycle loss will lead to very bad mapping results which didn't happen in my experiments. May I know how to get the results in the last column of Fig. 3? Thanks.

[fbxiang]

Removing the cycle loss means putting no restriction in the UV map, so it may or may not produce distorted UV maps. Different seed/initialization will result in different UV maps, and figure 3 just shows what I got after removing the cycle loss. I do not remember doing anything special, but since this UV map can be quite random, maybe doing the experiments using different random seeds/initializations would reproduce a more distorted map.