Update README.md
Browse files
README.md
CHANGED
|
@@ -130,4 +130,92 @@ configs:
|
|
| 130 |
- split: test
|
| 131 |
path: users/test-*
|
| 132 |
---
|
| 133 |
-
=====================
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 130 |
- split: test
|
| 131 |
path: users/test-*
|
| 132 |
---
|
| 133 |
+
=====================README====================
|
| 134 |
+
|
| 135 |
+
---
|
| 136 |
+
dataset_info:
|
| 137 |
+
description: |
|
| 138 |
+
Polyvore‑1000 is a curated fashion dataset derived from the Maryland Polyvore dataset (Polyvore‑U), enriched with local images and structured for personalized recommendation tasks.
|
| 139 |
+
citation:
|
| 140 |
+
- "@inproceedings{han2017learning, author = {Han, Xintong and Wu, Zuxuan and Jiang, Yu‑Gang and Davis, Larry S}, title = {Learning Fashion Compatibility with Bidirectional LSTMs}, booktitle = {ACM Multimedia}, year = {2017}}"
|
| 141 |
+
- "@inproceedings{lu2019learning, author = {Lu, Zhi and Hu, Yang and Jiang, Yunchao and Chen, Yan and Zeng, Bing}, title = {Learning Binary Code for Personalized Fashion Recommendation}, booktitle = {CVPR}, year = {2019}}"
|
| 142 |
+
- "@misc{polyvore_original, title = {Polyvore – social fashion platform (archived)}, howpublished = {https://www.polyvore.com}, note = {Inspiration for dataset}}"
|
| 143 |
+
|
| 144 |
+
dataset_creator:
|
| 145 |
+
- name: Xintong Han, Zuxuan Wu, Yu‑Gang Jiang, Larry S. Davis
|
| 146 |
+
affiliation: University of Maryland
|
| 147 |
+
contribution: Creators of the original Polyvore‑U splits used in the ACM MM 2017 study on fashion compatibility.
|
| 148 |
+
- name: Zhi Lu, Yang Hu, Yunchao Jiang, Yan Chen, Bing Zeng
|
| 149 |
+
affiliation: University of Electronic Science and Technology of China
|
| 150 |
+
contribution: Developed binary-code-based fashion recommendation algorithms using Polyvore‑U in CVPR 2019.
|
| 151 |
+
features:
|
| 152 |
+
items:
|
| 153 |
+
item_id: string
|
| 154 |
+
master_category: string
|
| 155 |
+
product_name: string
|
| 156 |
+
price: float
|
| 157 |
+
image: Image(path=str)
|
| 158 |
+
release_date: string
|
| 159 |
+
kits:
|
| 160 |
+
kit_id: string
|
| 161 |
+
kit_name: string
|
| 162 |
+
description: string
|
| 163 |
+
user_id: string
|
| 164 |
+
image: Image(path=str)
|
| 165 |
+
views: int
|
| 166 |
+
likes: int
|
| 167 |
+
date: string
|
| 168 |
+
users:
|
| 169 |
+
user_id: string
|
| 170 |
+
user_name: string
|
| 171 |
+
interactions:
|
| 172 |
+
user_id: string
|
| 173 |
+
item_id: string
|
| 174 |
+
interaction_type: string
|
| 175 |
+
date: string
|
| 176 |
+
---
|
| 177 |
+
|
| 178 |
+
# Polyvore‑1000 Dataset
|
| 179 |
+
|
| 180 |
+
Bienvenue ! Je suis **Waly NGOM**, docteur en mathématiques et passionné par l’intelligence artificielle. Ce dépôt contient **Polyvore‑1000**, un dataset conçu pour la recommandation personnalisée dans le domaine de la mode.
|
| 181 |
+
|
| 182 |
+
Polyvore‑1000 s’appuie sur les **splits Polyvore‑U** conçus par Han et al. (2017) et bénéficie du travail complémentaire de Lu et al. (CVPR 2019), qui ont apporté une approche innovante basée sur des codes binaires pour la recommandation efficace d’outfits :contentReference[oaicite:1]{index=1}.
|
| 183 |
+
|
| 184 |
+
## Structure des données
|
| 185 |
+
|
| 186 |
+
- **Splits disponibles** : `train`, `valid`, `test` (mêmes proportions que Polyvore‑U : 17 316 / 1 497 / 3 076 outfits).
|
| 187 |
+
- **Configurations** :
|
| 188 |
+
- `items` : données détaillées des items
|
| 189 |
+
- `kits` : informations sur chaque outfit
|
| 190 |
+
- `users` : identifiants synthétiques d’utilisateurs
|
| 191 |
+
- `interactions` : interactions entre utilisateurs et items (composition d’outfits, vues, likes)
|
| 192 |
+
|
| 193 |
+
### Images
|
| 194 |
+
|
| 195 |
+
Les images sont organisées dans `images/<kit_id>/` :
|
| 196 |
+
- `0.jpg` → image de l’outfit (kit)
|
| 197 |
+
- `1.jpg`, `2.jpg`, … → images correspondant aux items du kit, dans l’ordre des données JSON
|
| 198 |
+
|
| 199 |
+
## Utilisation
|
| 200 |
+
|
| 201 |
+
### Authentification Hugging Face
|
| 202 |
+
|
| 203 |
+
Dans un notebook ou script Python :
|
| 204 |
+
|
| 205 |
+
```python
|
| 206 |
+
from huggingface_hub import login
|
| 207 |
+
import os
|
| 208 |
+
|
| 209 |
+
login(token=os.getenv("HF_TOKEN"))
|
| 210 |
+
|
| 211 |
+
|
| 212 |
+
|
| 213 |
+
Pour charger ces datasets:
|
| 214 |
+
|
| 215 |
+
from datasets import load_dataset
|
| 216 |
+
|
| 217 |
+
items_ds = load_dataset("codewaly/polyvore1000", "items", split="train")
|
| 218 |
+
kits_ds = load_dataset("codewaly/polyvore1000", "kits", split="valid")
|
| 219 |
+
users_ds = load_dataset("codewaly/polyvore1000", "users")
|
| 220 |
+
interactions_ds = load_dataset("codewaly/polyvore1000", "interactions")
|
| 221 |
+
|