Research map / Label-efficient and robust learning / Recognition and long-tail
Attention-based Ensemble for Deep Metric Learning
Wonsik Kim, Bhavya Goyal, Kunal Chawla, Jungmin Lee, Keunjoo Kwon
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 divergence loss to ABE-8 (ablation) | ABE-8 (512) · CARS-196 | Recall@1 (%): 69.7 -> 85.2 | Improved |
| add divergence loss to 8-heads ensemble (ablation) | 8-heads · CARS-196 | Recall@1 (%): 76.1 -> 76.0 | No clear effect |
| add attention modules to 8-heads (ablation) | 8-heads · CARS-196 | Recall@1 (%): 76.1 -> 79.7 | Improved |
| share global embedding G across learners and add divergence loss on top of attention (ablation) | 8-heads+att · CARS-196 | Recall@1 (%): 79.7 -> 85.2 | Improved |
| use shared G with separate S_m (8-tails) instead of 8-heads (ablation) | 8-heads · CARS-196 | Recall@1 (%): +5.0% | Improved |
| add attention modules to 8-tails (ablation) | 8-tails · CARS-196 | Recall@1 (%): 81.1 -> 85.2 | Improved |
| increase number of attention learners M from 1 to 8 (ablation) | ABE-M (512) · CARS-196 | Recall@1 (%): 67.3 -> 85.2 | Improved |
| replace 4-heads ensemble with ABE-4 attention ensemble | 4-heads (512) · CARS-196 | Recall@1 (%): 76.6 -> 82.5 | Improved |
| increase M with fixed 64-d individual learner embedding (ablation) | ABE-M individual learners · CARS-196 | Recall@1 (%): 65.9 -> 75.0 | Improved |
| increase M in M-heads from 4 to 8 heads (ablation) | M-heads · CARS-196 | Recall@1 (%): 76.6 -> 76.1 | No clear effect |
| use 5 inception blocks in attention module (4a to 4e) (ablation) | ABE-8 · CARS-196 | Recall@1 (%) | Improved |
| branch attention at pool3 (ablation) | ABE-8 · CARS-196 | Recall@1 (%) | Improved |
Closest papers by meaning
- Deep Metric Learning with BIER: Boosting Independent Embeddings Robustly (TPAMI 2018 · similarity 0.90)
- Deep Randomized Ensembles for Metric Learning (ECCV 2018 · similarity 0.88)
- Towards Improved and Interpretable Deep Metric Learning via Attentive Grouping (TPAMI 2022 · similarity 0.87)
- BIER - Boosting Independent Embeddings Robustly (ICCV 2017 · similarity 0.86)
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.86)
- Unsupervised Deep Metric Learning with Transformed Attention Consistency and Contrastive Clustering Loss (ECCV 2020 · 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.
