Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Intra-Batch Connections for Deep Metric Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 message passing over all mini-batch samples during training (ablation) | ResNet50 metric-learning model · CUB-200-2011 | Recall@1: 67.5 -> 70.3; 2.8pp | Helped |
| Add message passing over all mini-batch samples during training (ablation) | ResNet50 metric-learning model · Cars196 | NMI: 68.7 -> 74.8; 6.1pp | Helped |
| Add the auxiliary cross-entropy loss to the backbone (ablation) | Intra-Batch Connections · CUB-200-2011 | Recall@1: 68.1 -> 70.3; 2.2pp | Helped |
| Add the auxiliary cross-entropy loss to the backbone (ablation) | Intra-Batch Connections · Cars196 | NMI: 72.1 -> 74.8; 2.7pp | Helped |
| Use MPN-refined embeddings at test time (ablation) | Intra-Batch Connections · Cars196 | NMI: 74.8 -> 76.2; 1.4pp | Helped |
| Use reciprocal kNN batches for test-time message passing (ablation) | Intra-Batch Connections · CUB-200-2011 | Recall@1: 70.3 -> 70.8; 0.5pp | Helped |
| Construct test-time batches from ordinary nearest neighbors (ablation) | Intra-Batch Connections · CUB-200-2011 | Recall@1: 70.3 -> 62.1 | Hurt |
| Use five independently trained networks as an ensemble (ablation) | Intra-Batch Connections · Cars196 | Recall@1: 88.1 -> 91.5; 3.4pp | Helped |
| Increase embedding dimension from 512 to 1024 (ablation) | Intra-Batch Connections · Stanford Online Products | Recall@1: 81.4 -> 81.4 | No clear effect |
| Increase embedding dimension from 512 to 1024 (ablation) | Intra-Batch Connections · In-Shop Clothes | Recall@1: 92.8 -> 91.7 | Hurt |
| Use four message-passing steps (ablation) | Intra-Batch Connections · CUB-200-2011 | Recall@1 | Hurt |
| Distill MPN class probabilities into the backbone (ablation) | Intra-Batch Connections · CUB-200-2011 | Recall@1: -4.6pp | Hurt |
Closest papers by meaning
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.89)
- DarkRank: Accelerating Deep Metric Learning via Cross Sample Similarities Transfer (AAAI 2018 · similarity 0.88)
- Deep Metric Learning via Lifted Structured Feature Embedding (CVPR 2016 · similarity 0.88)
- Ranked List Loss for Deep Metric Learning (CVPR 2019 · similarity 0.87)
- Deep Metric Learning with Self-Supervised Ranking (AAAI 2021 · similarity 0.87)
- Learning Metrics From Teachers: Compact Networks for Image Embedding (CVPR 2019 · similarity 0.87)
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.