Now I am running C3D model for the action similarity labeling (2nd task in your paper).
For some similarity index, we need to compute division (e.g. in x1*(log(x1/x2)), and the output from layers could be zero. Therefore, the divisor is zero, leading to the problem that the similarity is NAN.
Did you meet such problem? If so, how did you address this case? Many thanks.
Hi, Du Tran
Thanks for your code.
Now I am running C3D model for the action similarity labeling (2nd task in your paper).
For some similarity index, we need to compute division (e.g. in x1*(log(x1/x2)), and the output from layers could be zero. Therefore, the divisor is zero, leading to the problem that the similarity is NAN.
Did you meet such problem? If so, how did you address this case? Many thanks.