Back to datasets
Dataset assetOpen Source CommunityData AnalysisFashion Matching
Polyvore Dataset
This dataset includes 21,889 outfits from polyvore.com for training, validation, and testing. Each outfit contains name, view count, clothing items, image URL, likes, upload date, and description. The dataset also provides question‑answer pairs for evaluating fill‑in‑the‑blank fashion recommendation tasks, as well as data for fashion compatibility prediction tasks.
Source
github
Created
Aug 26, 2017
Updated
May 8, 2024
Signals
850 views
Availability
Linked source ready
Overview
Dataset description and usage context
Dataset Overview
Dataset Name
- Polyvore Dataset
Source
- Derived from the ACM MM17 paper "Learning Fashion Compatibility with Bidirectional LSTMs".
Content
- Polyvore outfits: 21,889 outfits (17,316 train, 1,497 validation, 3,076 test).
- JSON structure: Each outfit includes name, view count, list of clothing items, image URL, likes, upload date, URL, and description.
- Clothing item info: index, name, price, likes, image URL, category ID.
- Image URLs: Currently dead; can be retrieved via the Kaggle page.
- Category ID mapping: Provided as a separate file.
Tasks
- Fill‑in‑the‑blank Fashion Recommendation: Includes evaluation data with question, answer, and blank position.
- Fashion Compatibility Prediction: About 7,000 outfits for compatibility evaluation.
Download
- Available via Google Drive and Kaggle.
Version
- A cleaned version named Cleaned Maryland is provided.
Contact
- Author: Xintong Han (xintong@umd.edu)
Citation
- Follow the citation format in the README.
Need downstream help?
Pair the dataset with AI analysis and content workflows.
Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.