DRIT++ performs surprisingly poor on edges2shoes

[doantientai]

Congrats Lee and your team for your great work!

I have tried the method on difference datasets and it worked very well. However, edges2shoes is an exception. I have in the attachment the result after 14 epochs of training (~40 hours).

I am wondering if edges is the problem? Have anyone tried training the method with a similar dataset?

Thank you!