Research map / Label-efficient and robust learning / Recognition and long-tail
Contrastive Bayesian Analysis for Deep Metric Learning
Shichao Kan, Zhiquan He, Yigang Cen, Yang Li, Vladimir M. Mladenovic, Zhihai He
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 log averaging of conditional probabilities instead of identity (ablation) | CBML · CUB | Recall@1: 69.2 -> 69.9 | Improved |
| use log averaging instead of sqrt (ablation) | CBML · SOP | Recall@1: 78.6 -> 79.9 | Improved |
| use log averaging instead of sqrt (ablation) | CBML · In-Shop | Recall@1: 89.7 -> 92.3 | Improved |
| use log averaging instead of sqrt (ablation) | CBML · CUB | Recall@1: 70.0 -> 69.9 | Hurt |
| add metric variance constraint (lambda 0.0 -> 1.0) (ablation) | CBML · CUB | Recall@1: 63.8 -> 69.9 | Improved |
| increase embedding size from 64 to 512 (ablation) | CBML · CUB | Recall@1: 64.3 -> 69.9 | Improved |
| increase embedding size from 512 to 1024 (ablation) | CBML · CUB | Recall@1: 69.9 -> 69.0 | Hurt |
| increase embedding size from 512 to 1024 (ablation) | CBML · In-Shop | Recall@1: 92.3 -> 92.5 | No clear effect |
| use CBML instead of Margin-style best prior (MS) at 64-d | CBML · CUB | Recall@1: 57.4 -> 64.3 | Improved |
| use CBML instead of MS at 64-d | CBML · Cars | Recall@1: 77.3 -> 83.2 | Improved |
| use CBML instead of A-BIER with GoogLeNet | CBML · In-Shop | Recall@1: 83.1 -> 88.7 | Improved |
| use CBML instead of LoOp+MS with GoogLeNet | CBML · SOP | Recall@1: 76.6 -> 74.8 | Hurt |
Closest papers by meaning
- Contrastive Learning with Hard Negative Samples (ICLR 2021 · similarity 0.86)
- Bayesian Metric Learning for Uncertainty Quantification in Image Retrieval (NeurIPS 2023 · similarity 0.85)
- Embedding Transfer With Label Relaxation for Improved Metric Learning (CVPR 2021 · similarity 0.85)
- It Takes Two to Tango: Mixup for Deep Metric Learning (ICLR 2022 · similarity 0.85)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.84)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 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.
