How exactly spheric_pose defined?

[povolann]

Hello,

I have read your comments on this problem: #16 and it seems very useful. I managed to get better results with modifying trans_t or spheric_poses, but can you explain more or share how exactly are these variables defined? I know a bit (like trans_t is translation wrt poses centre) from your previous comments, but I am still struggling to find a more detailed understanding of these.

Also, if I changed these parameters, do I have to re-train the whole thing? Because I went through the code, I think that these changes only affect the final output of eval.py, but not the training result.

Thank you and have a nice day!