Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Type-Aware Embeddings for Fashion Compatibility
Mariya I. Vasileva, Bryan A. Plummer, Krishna Dusad, Shreya Rajpal, Ranjitha Kumar, David Forsyth
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn type-specific projections (CSN T1:1) instead of one shared embedding | SiameseNet · Maryland Polyvore (category-aware negatives) | FITB accuracy: 54.4 -> 57.9 | Improved |
| add visual-semantic embedding (VSE) and similarity (Sim) constraints (ablation) | CSN T1:1 · Maryland Polyvore (category-aware negatives) | FITB accuracy: 57.9 -> 59.0 | Improved |
| add learned generalized distance metric (ablation) | CSN T1:1 + VSE + Sim · Maryland Polyvore (category-aware negatives) | FITB accuracy: 59.0 -> 61.0 | Improved |
| increase embedding size from 64-D to 512-D (ablation) | CSN T1:1 + VSE + Sim + Metric · Maryland Polyvore (category-aware negatives) | FITB accuracy: 61.0 -> 65.0 | Improved |
| reduce embedding size from 64-D to 32-D (ablation) | CSN T1:1 + VSE + Sim + Metric · Maryland Polyvore (category-aware negatives) | FITB accuracy: 61.0 -> 55.7 | Hurt |
| type-specific projections without VSE/Sim/Metric under random negatives | SiameseNet · Maryland Polyvore (all negatives) | FITB accuracy: 54.2 -> 51.6 | Hurt |
| add learned metric on top of VSE and Sim under random negatives (ablation) | CSN T1:1 · Maryland Polyvore (all negatives) | FITB accuracy: 51.5 -> 86.1 | Improved |
| share each projection across four type pairs (T4:1) instead of one per pair (ablation) | CSN + VSE + Sim + Metric · Polyvore Outfits | FITB accuracy: 55.3 -> 55.1 | Hurt |
| type-specific projections instead of single embedding | SiameseNet · Polyvore Outfits | Compat. AUC: 0.81 -> 0.83 | Improved |
| use cosine distance instead of learned metric (ablation) | CSN T1:1 + VSE + Sim · Polyvore Outfits | FITB accuracy: 55.3 -> 54.8 | Hurt |
| use fully connected projection instead of diagonal projection (ablation) | CSN T1:1 + VSE + Sim + Metric · Polyvore Outfits | FITB accuracy: 55.3 -> 56.6 | Improved |
| increase embedding size from 32-D to 512-D (ablation) | CSN T1:1 + VSE + Sim + Metric · Polyvore Outfits-D | FITB accuracy: 53.2 -> 55.2 | Improved |
Closest papers by meaning
- Context-Aware Visual Compatibility Prediction (CVPR 2019 · similarity 0.90)
- Learning to Match on Graph for Fashion Compatibility Modeling (AAAI 2020 · similarity 0.90)
- TransNFCM: Translation-Based Neural Fashion Compatibility Modeling (AAAI 2019 · similarity 0.89)
- Learning the Latent "Look": Unsupervised Discovery of a Style-Coherent Embedding From Fashion Images (ICCV 2017 · similarity 0.89)
- Easy Regional Contrastive Learning of Expressive Fashion Representations (NeurIPS 2024 · similarity 0.87)
- Deciphering Compatibility Relationships with Textual Descriptions via Extraction and Explanation (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
