The code will be given here soon after the paper is accepted.

The following figure gives a visualization of the endmember optimization of our UNTR on the Jasper Ridge dataset during training, using only random initialization and not relying on any pre-established endmember estimation results, such as VCA and N-Finder.