About the no_dup data

[HITKevin]

trafficstars

Hi , I noticed the train/test/valid_no_dup.json in dataset polyvore is different from the previous dataset released by paper Learning Type-Aware Embeddings for Fashion Compatibility. I want to know how you preprocessed the dataset

[gcucurull]

Hello,

We haven't changed those files, out script just downloads them from here, which is the link the authors of the dataset provide in https://github.com/xthan/polyvore-dataset.

[HITKevin]

Thanks for your explanation. I just noticed Polyvore-dataset and Polyvore-outfits are two different datasets. Have you ever tested your method on Polyvore-outfits or compared with paper Learning Type-Aware Embeddings for Fashion Compatibility in any ways?

[HITKevin]

[gcucurull]

We have only used the Maryland Polyvore dataset from [12], not the Polyvore-outfits dataset they introduced in the paper. If I remember correclty, it wasn't released at the time.