Thanks for share this valuable work! My concern is that whether the 36.7M high quality triplets training dataset will be open for downloading? Thanks~