bismex / hicmd Goto Github PK

Hi,
Thanks for making the code available.
I have a query in the KL divergence loss term, where the KL loss between ID excluded attribute representation and the Gaussian distribution is tried to be minimised.
Actually I don't understand why is this loss written this way in code, some indices are chosen at att_ex_idx and values at those indices are given to compute_kl method. First, what does those indices signify and second in the compute_kl method, it just computes the mean of the square of the values.
I also want to add a smiilar loss in my work. Please redirect me to some resource which gives some explanation about the closed form solution for KL divergence between the latent variable and Gaussian distribution.

Thanks a lot in advance!

Dear scholar,
Could you give me the link about the supplementary materials that was mentioned in your paper? And maybe the introduction of CE loss part is very brief. I want to ask you whether feature fi(from di com) is followed by a classifier to get the CE loss. And how to decide the number of nodes in the classifier? I guess the number of nodes in classifier maybe the number of ids in the training set such as 206 or 395.

trainer.py line 88-112
self.att_pose_idx
self.att_illum_idx
define twice

Excuse me.
If my GPU memory is 8， how can I modify it to make the code run？
Thank you！！

The RegDB website requires an application.
Could you share RegDB dataset?
I will be very grateful to you.
my mail is: [email protected]
Thanks!

Nice work! Where can I get the pretrained model of SYSU?