Research map / Graph learning / Graph algorithms and clustering
Learning Better Visual Data Similarities via New Grouplet Non-Euclidean Embedding
Yanfu Zhang, Lei Luo, Wenhan Xian, Heng Huang
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 Poincare ball layer instead of Euclidean layer (ablation) | Proposed512 · CUB-200-2011 | Recall@1: 74.8 -> 75.4 | Improved |
| use Poincare ball layer instead of Euclidean layer (ablation) | ProxyAnchor512 · CUB-200-2011 | Recall@1: 68.4 -> 68.8 | Improved |
| use Poincare ball layer instead of Euclidean layer (ablation) | ProxyGML512 · CUB-200-2011 | Recall@1: 66.6 -> 67.1 | Improved |
| use hyperboloid layer instead of Euclidean layer (ablation) | Proposed512 · CUB-200-2011 | Recall@1: 74.8 -> 73.6 | Hurt |
| use hyperboloid layer instead of Euclidean layer (ablation) | ProxyGML512 · CUB-200-2011 | Recall@1: 66.6 -> 65.9 | Hurt |
| use hyperboloid layer instead of Euclidean layer (ablation) | ProxyAnchor512 · CUB-200-2011 | Recall@1: 68.4 -> 68.1 | Hurt |
| increase batch size from 64 to 180 | ProxyAnchor512 · CUB-200-2011 | Recall@1: 67.1 -> 69.0 | Improved |
| apply grouplet optimal transport at batch 64 (16 x 4) instead of ProxyAnchor (ablation) | Proposed512 · CUB-200-2011 | Recall@1: 67.1 -> 75.4 | Improved |
| increase number of grouplets from 8 to 16 (size 4) (ablation) | Proposed512 · CUB-200-2011 | Recall@1: 73.6 -> 75.4 | Improved |
| increase number of grouplets from 16 to 32 (size 4) (ablation) | Proposed512 · CUB-200-2011 | Recall@1: 75.4 -> 75.5 | No clear effect |
| vary grouplet size at similar batch size (16x4, 8x8, 12x6) (ablation) | Proposed512 · CUB-200-2011 | Recall@1: 75.4, 74.9, 74.7 | No clear effect |
| vary Poincare ball radius c (ablation) | Proposed512 · CUB-200-2011 | Recall@1 | No clear effect |
Closest papers by meaning
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.89)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.88)
- Simultaneous Similarity-based Self-Distillation for Deep Metric Learning (ICML 2021 · similarity 0.87)
- DiVA: Diverse Visual Feature Aggregation for Deep Metric Learning (ECCV 2020 · similarity 0.87)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · similarity 0.86)
- Volume-Aware Distance for Robust Similarity Learning (ICML 2025 · similarity 0.84)
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.
