Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Similarity Conditions Without Explicit Supervision
Reuben Tan, Mariya I. Vasileva, Kate Saenko, Bryan A. Plummer
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 |
|---|---|---|---|
| use dynamic condition weight branch with 5 masks instead of a single condition (ablation) | SCE-Net · Polyvore Outfits val | FITB accuracy (%): 53.2 -> 62.1 | Improved |
| use 5 instead of 10 condition masks (ablation) | SCE-Net · Polyvore Outfits val | FITB accuracy (%): 60.8 -> 62.1 | Improved |
| use 5 instead of 20 condition masks (ablation) | SCE-Net · Polyvore Outfits val | compatibility AUC: 0.89 -> 0.92 | Improved |
| use SCE-Net with no test-time supervision instead of type-aware strongly supervised embeddings | outfit compatibility · Polyvore Outfits test | FITB accuracy (%): 55.3 -> 61.6 | Improved |
| use SCE-Net instead of type-aware embeddings | outfit compatibility · Maryland Polyvore test | compatibility AUC: 0.90 -> 0.90 | No clear effect |
| use SCE-Net instead of Siamese Net on unseen categories | fill-in-the-blank · Scarves unseen category | FITB accuracy (%): 46.62 -> 59.46 | Improved |
| use SCE-Net instead of Siamese Net on unseen categories | fill-in-the-blank · Accessories unseen category | FITB accuracy (%): 50.82 -> 56.55 | Improved |
| condition weight branch on labels (text) instead of visuals (ablation) | SCE-Net · Polyvore Outfits test | FITB accuracy (%): 61.6 -> 60.8 | No clear effect |
| add label text features to visual features in the weight branch (ablation) | SCE-Net · Polyvore Outfits test | FITB accuracy (%): 61.6 -> 61.5 | No clear effect |
| use SCE-Net with 4 masks and no condition label instead of strongly supervised CSN | triplet prediction · UT-Zappos50K | error rate (%): 10.73% -> 7.53% | Improved |
| use 4 learned masks instead of 2 (ablation) | SCE-Net · UT-Zappos50K | error rate (%): 11.12% -> 7.53% | Improved |
| use 200k instead of 100k training triplets (ablation) | SCE-Net · UT-Zappos50K | error rate (%): 11.37% -> 8.29% | Improved |
Closest papers by meaning
- Conditional Similarity Networks (CVPR 2017 · similarity 0.92)
- Generalized Conditional Similarity Learning via Semantic Matching (TPAMI 2025 · similarity 0.89)
- Effectively Leveraging Attributes for Visual Similarity (ICCV 2021 · similarity 0.88)
- Learning to Align Generative Appearance Priors for Fine-grained Image Retrieval (NeurIPS 2026 · similarity 0.86)
- Weakly-Supervised Contrastive Learning for Imprecise Class Labels (ICML 2025 · similarity 0.86)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.86)
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.
