Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Metric Learning with BIER: Boosting Independent Embeddings Robustly
Michael Opitz, Georg Waltner, Horst Possegger, Horst Bischof
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 |
|---|---|---|---|
| add boosting-based reweighting on top of Activation Loss initialization (ablation) | BIER 96-160-256 · CUB-200-2011 | Recall@1: 53.93 -> 55.33 | Improved |
| add Activation Loss initialization with split groups and no boosting (ablation) | Baseline-512 · CUB-200-2011 | Recall@1: 51.76 -> 53.93 | Improved |
| use groups of unequal size 96-160-256 instead of 170-171-171 (ablation) | BIER · CUB-200-2011 | Recall@1: 54.76 -> 55.33 | Improved |
| apply BIER to triplet loss (ablation) | Triplet-512 · CUB-200-2011 | Recall@1: 50.12 -> 53.31 | Improved |
| apply BIER to contrastive loss (ablation) | Contrastive-512 · CUB-200-2011 | Recall@1: 50.62 -> 53.8 | Improved |
| apply BIER to binomial deviance loss (ablation) | Binomial-Deviance-512 · CUB-200-2011 | Recall@1: 51.76 -> 55.33 | Improved |
| increase groups from 4 to 5 at 512-d embedding (ablation) | BIER · CUB-200-2011 | Recall@1: 55.62 -> 54.9 | Hurt |
| apply BIER to 1024-d embedding (ablation) | Baseline-1024 · CUB-200-2011 | Recall@1: 52.89 -> 55.99 | Improved |
| apply BIER to 384-d embedding (ablation) | Baseline-384 · CUB-200-2011 | Recall@1: 51.57 -> 54.66 | Improved |
| use Activation Loss initialization instead of Glorot random initialization (ablation) | BIER · CUB-200-2011 | Recall@1: 54.41 -> 55.33 | Improved |
| use Activation Loss initialization instead of orthogonal initialization (ablation) | BIER · CUB-200-2011 | Recall@1: 54.58 -> 55.33 | Improved |
| raise learning rate to 1e-5 without auxiliary loss (ablation) | BIER · CUB-200-2011 | Recall@1: 55.3 -> 52.3 | Hurt |
Closest papers by meaning
- BIER - Boosting Independent Embeddings Robustly (ICCV 2017 · similarity 0.98)
- Deep Randomized Ensembles for Metric Learning (ECCV 2018 · similarity 0.91)
- Attention-based Ensemble for Deep Metric Learning (ECCV 2018 · similarity 0.90)
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.89)
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.89)
- Deep Metric Learning with Adaptively Composite Dynamic Constraints (TPAMI 2023 · similarity 0.88)
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.
