lzcn/outfit-datasets

Outfit Datasets

Processed fashion outfit datasets: item images, pre-extracted ResNet-34 features (LMDB), and outfit tuple splits. Ready to use, no preprocessing required.

Layout

polyvore-u/                      # PolyVore-U
├── features/images/data.mdb     # item images (LMDB, keyed by item file name)
├── features/resnet34/data.mdb   # ResNet-34 features
└── original/tuples_{519,630,53,32}/   # personalized / new-user splits

iqon-3000/                       # IQON3000
├── features/images_l/data.mdb   # large-resolution images
├── features/images_m/data.mdb   # medium-resolution images
├── features/resnet34/data.mdb   # ResNet-34 features
└── iqon-{550,58}/               # personalized / new-user splits

maryland-polyvore/               # Maryland-PolyVore (non-personalized)
├── features/{images,resnet34}/data.mdb
└── original/

polyvore-outfits/                # PolyVore (UIUC)
├── features/{images,resnet34}/data.mdb
├── disjoint/                    # disjoint FITB split
└── nondisjoint/                 # non-disjoint FITB split

Tuple Files

Each split folder contains CSV text files (train_pos, train_neg, valid_pos, valid_neg, test_pos, test_neg, test_fitb, test_pos_fitb, test_neg_fitb, plus items.json and retrieval where applicable). One outfit per line:

uidx,size,item_1,...,item_K,type_1,...,type_K

padded with -1 to a fixed number of slots K. IQON tuples omit the size column; outfit size is inferred from the valid item IDs.

Usage

hf download lzcn/outfit-datasets --repo-type dataset --local-dir data --exclude "*/features/images*/*"

Large files use Git LFS; lock.mdb files are auto-generated by LMDB and can be ignored.

Datasets are preprocessed from PolyVore-U, IQON3000, Maryland-PolyVore, and PolyVore (UIUC) with lzcn/outfit-datasets.

Dataset sources

Please cite the original dataset papers when using the data.

Dataset	Original paper
Polyvore-U	Lu et al., Learning Binary Code for Personalized Fashion Recommendation, CVPR 2019
IQON-3000	Song et al., GP-BPR: Personalized Compatibility Modeling for Clothing Matching, ACM MM 2019
Maryland Polyvore	Han et al., Learning Fashion Compatibility with Bidirectional LSTMs, ACM MM 2017
Polyvore Outfits	Vasileva et al., Learning Type-Aware Embeddings for Fashion Compatibility, ECCV 2018