Reproduce “Fashionpedia: Ontology, Segmentation, and an Attribute Localization Dataset” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Menglin Jia, Mengyun Shi, Mikhail Sirotenko, Yin Cui, Claire Cardie, Bharath Hariharan, Hartwig Adam, Serge Belongie
Researchers from Cornell University and Google Research developed Fashionpedia, a dataset and ontology enabling a new computer vision task: instance segmentation with attribute localization for fashion items. This resource provides exhaustive segmentation masks for garments and their parts, along with localized, expert-annotated fine-grained attributes, revealing a substantial performance gap for models tackling this combined challenge.
Key insights
- A unified, expert-curated ontology is essential for structuring complex domain knowledge in fine-grained visual tasks.
- Exhaustive, part-level segmentation and localized, expert-annotated attributes provide a richer, more structured visual representation.
- Combining instance segmentation with attribute localization is considerably more challenging than segmentation alone, indicating significant room for model development.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.