BarnesHutTsne does not produce expected output

Hello. I’m using BarnesHutTsne, but the result does not look like as expected. I tried MNist data as well as my own data samples.

The T-SNE implementation of Leif Jonsson ( just works. But DL4J implementation always results in data arranged as a donut. The resulting data vector is not even a t-SNE ball.

Does the BarnesHutTsne implementation really work?