Open benoriol opened 8 months ago
I have the same question regarding the clip direction consistency similarity.
You just have to change the inputs to the function when you call it. So if you are doing the directional score, input consecutive frames, and if you are doing similarity score, input the two frames from the separate videos
Originally posted by @ayaanzhaque in https://github.com/ayaanzhaque/instruct-nerf2nerf/issues/82#issuecomment-1852544207
From this response, the similarity score would be (C(e_i)) - C(o_i)) * (CT(e_i)) - CT(o_i))
Which is also different
Hello,
Seems there might be an error in the formulation of your clip direction consistency similarity, according to #28 ? It is written down as (C(e_i) - C(o_i)) * (C(e_i+1) - C(e_i))
Did you mean to write (C(o_i+1)) - C(o_i)) (C(e_i+1) - C(e_i)) or maybe (C(e_i)) - C(o_i)) (C(e_i+1) - C(o_i+1)) ?
In the previous issue you said it is the first one, but I don't know if that is actually correct? I would really appreciate if you could confirm which one of the two is it.
Thanks!