15 million fashion-related data points organized for set-to-set matching tasks across multiple distribution shifts. The dataset focuses on fashion items and their relationships within sets to evaluate model performance under varying data distributions.
Use Cases
- Evaluate the performance of set-to-set matching models using the provided distribution shift partitions
- Train fashion recommendation engines that predict item compatibility within sets
- Research domain adaptation techniques for fashion data using the shift-specific subsets
Strengths
- 15 million data points focused on fashion items and set-based relationships
- Explicitly partitioned to represent multiple types of distribution shifts
- Designed for set-to-set matching tasks rather than standard classification or regression