codewaly/polyvore1000

=====================README====================

Polyvore-1000 Dataset

Welcome! I am Waly NGOM, PhD in Mathematics and passionate about Artificial Intelligence. This repository contains Polyvore-1000, a dataset designed for personalized recommendation in the fashion domain.

Polyvore-1000 builds upon the Polyvore-U splits introduced by Han et al. (2017) and benefits from the complementary work of Lu et al. (CVPR 2019), who proposed an innovative binary-code based approach for efficient outfit recommendation.

Data Structure

a. Available splits: train, valid, test (same proportions as Polyvore-U: 17,316 / 1,497 / 3,076 outfits).

b. Configurations:

items: detailed item data

kits: information on each outfit

users: synthetic user identifiers

interactions: interactions between users and items (outfit composition, views, likes)

user_profiles: aggregated user interaction profiles

Images

Images are organized in images//:

• 0.jpg → outfit (kit) image

• 1.jpg, 2.jpg, … → images corresponding to the items of the kit, in the order given by the JSON data

Hugging Face Authentication

In a notebook or Python script:

from huggingface_hub import login import os

login(token=os.getenv("HF_TOKEN"))

Usage

To load these datasets:

from datasets import load_dataset

items_ds = load_dataset("codewaly/polyvore1000", "items", split="train")

kits_ds = load_dataset("codewaly/polyvore1000", "kits", split="train")

users_ds = load_dataset("codewaly/polyvore1000", "users", split="train")

interactions_ds = load_dataset("codewaly/polyvore1000", "interactions", split="train")

user_profiles_ds = load_dataset("codewaly/polyvore1000", "user_profiles", split="train")

References

1. Han, X., et al. (2017). Learning Fashion Compatibility with Bidirectional LSTMs. ACM Multimedia.

2. Lu, Z., et al. (2019). Learning Binary Code for Personalized Fashion Recommendation. CVPR.