Research map / Label-efficient and robust learning / Recognition and long-tail
Identifying Ambiguous Similarity Conditions via Semantic Matching
Han-Jia Ye, Yi Shi, De-Chuan Zhan
Check 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 DiscoverNet Reg instead of LSN | WS-CSL on UT-Zappos-50k with pretraining · UT-Zappos-50k, w/ pretrain | GR accuracy: 76.26 -> 77.84 | Improved |
| Use DiscoverNet Reg instead of LSN | WS-CSL on UT-Zappos-50k with pretraining · UT-Zappos-50k, w/ pretrain | OT accuracy: 75.90 -> 77.68 | Improved |
| Use DiscoverNet Set instead of LSN | WS-CSL on UT-Zappos-50k without pretraining · UT-Zappos-50k, w/o pretrain | GR accuracy: 71.49 -> 74.67 | Improved |
| Use DiscoverNet Set instead of LSN | WS-CSL on UT-Zappos-50k without pretraining · UT-Zappos-50k, w/o pretrain | OT accuracy: 68.49 -> 74.13 | Improved |
| Use DiscoverNet Set instead of LSN | WS-CSL on Celeb-A · Celeb-A | GR accuracy: 70.29 -> 78.45 | Improved |
| Use DiscoverNet Set instead of LSN | WS-CSL on Celeb-A · Celeb-A | OT accuracy: 69.89 -> 77.98 | Improved |
| Use DiscoverNet Set instead of LSN | WS-CSL on Celeb-A† · Celeb-A† | GR accuracy: 57.02 -> 64.57 | Improved |
| Use DiscoverNet Set instead of LSN | WS-CSL on Celeb-A† · Celeb-A† | OT accuracy: 56.33 -> 63.98 | Improved |
| Increase projection count from 2 to 8 in DiscoverNet Set (ablation) | DiscoverNet Set on Celeb-A with pretraining · Celeb-A, w/ pretrain | OT accuracy: 77.74 -> 78.81 | Improved |
| Increase projection count from 8 to 10 in DiscoverNet Set (ablation) | DiscoverNet Set on Celeb-A with pretraining · Celeb-A, w/ pretrain | OT accuracy: 78.81 -> 78.68 | Hurt |
| Increase projection count from 2 to 8 in DiscoverNet Reg (ablation) | DiscoverNet Reg on Celeb-A with pretraining · Celeb-A, w/ pretrain | OT accuracy: 71.82 -> 74.07 | Improved |
| Increase projection count from 8 to 10 in DiscoverNet Reg (ablation) | DiscoverNet Reg on Celeb-A with pretraining · Celeb-A, w/ pretrain | OT accuracy: 74.07 -> 73.77 | Hurt |
Closest papers by meaning
- Generalized Conditional Similarity Learning via Semantic Matching (TPAMI 2025 · similarity 0.92)
- Conditional Similarity Networks (CVPR 2017 · similarity 0.88)
- Learning Similarity Conditions Without Explicit Supervision (ICCV 2019 · similarity 0.88)
- Weakly-Supervised Contrastive Learning for Imprecise Class Labels (ICML 2025 · similarity 0.86)
- SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery (NeurIPS 2025 · similarity 0.82)
- MERIT: Multilingual Semantic Retrieval with Interleaved Multi-Condition Query (NeurIPS 2025 · similarity 0.82)
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.
