Hi,
I read your paper and it was very insightful. I was curious about your dataset. However, when I downloaded fashion.json and computed the statistics my numbers are not matching with those of the paper.
I am getting the following numbers:
#scene : 43515 ( In paper it was 47,739 )
#product : 38111
#scene-product pair: 72198
I have used scene_bbox as a primary key for computing the number of unique scenes.
Regards