Deep Metric Learning for Cross-Domain Fashion Instance Retrieval

Abstract

The goal of this paper is to find an effective method to retrieve an image with a fashion instance from one domain based on a similar fashion instance image from a different domain. Where existing works focus on retrieving relevant shop images based on a consumer instance, we introduce the reverse task and treat both tasks equally in our training setup. We use several deep metric learning techniques to get baseline scores for these tasks on the DeepFashion2 dataset and we show how ensemble methods can be used to boost the performance.

Cite

Text

Ibrahimi et al. "Deep Metric Learning for Cross-Domain Fashion Instance Retrieval." IEEE/CVF International Conference on Computer Vision Workshops, 2019. doi:10.1109/ICCVW.2019.00390

Markdown

[Ibrahimi et al. "Deep Metric Learning for Cross-Domain Fashion Instance Retrieval." IEEE/CVF International Conference on Computer Vision Workshops, 2019.](https://mlanthology.org/iccvw/2019/ibrahimi2019iccvw-deep/) doi:10.1109/ICCVW.2019.00390

BibTeX

@inproceedings{ibrahimi2019iccvw-deep,
  title     = {{Deep Metric Learning for Cross-Domain Fashion Instance Retrieval}},
  author    = {Ibrahimi, Sarah and van Noord, Nanne and Geradts, Zeno J. M. H. and Worring, Marcel},
  booktitle = {IEEE/CVF International Conference on Computer Vision Workshops},
  year      = {2019},
  pages     = {3165-3168},
  doi       = {10.1109/ICCVW.2019.00390},
  url       = {https://mlanthology.org/iccvw/2019/ibrahimi2019iccvw-deep/}
}