Research map / Label-efficient and robust learning / Recognition and long-tail
DiVA: Diverse Visual Feature Aggregation for Deep Metric Learning
Timo Milbich, Karsten Roth, Homanga Bharadhwaj, Samarth Sinha, Yoshua Bengio, Björn Ommer, Joseph Paul Cohen
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 all four tasks (shared, intra-class, DaNCE) to margin-loss discriminative training (ablation) | Margin loss, Inception-BN · CUB200-2011 | Recall@1: 63.6 -> 66.4 | Improved |
| add all four tasks (ablation) | Margin loss, Inception-BN · CARS196 | Recall@1: 79.4 -> 83.1 | Improved |
| add all four tasks (ablation) | Margin loss, Inception-BN · SOP | Recall@1: 76.6 -> 77.5 | Improved |
| add the intra-class task to D & Da & S (ablation) | DiVA, Inception-BN · CARS196 | Recall@1: 82.6 -> 83.1 | Improved |
| add the shared task to D & Da & I (ablation) | DiVA, Inception-BN · SOP | Recall@1: 75.8 -> 77.5 | Improved |
| add embedding decorrelation between tasks (ablation) | DiVA, Inception-BN · CUB200-2011 | Recall@1: 65.6 -> 66.4 | Improved |
| share the feature encoder across tasks instead of training separate models (ablation) | DiVA · CUB200-2011 | Recall@1: 48.7 -> 66.4 | Improved |
| use multiple diverse tasks without decorrelation versus margin baseline (ablation) | margin loss baseline · CUB200-2011 | Recall@1: 63.6 -> 65.6 | Improved |
| use distance-adapted NCE (DaNCE) instead of standard NCE as extra task (ablation) | discriminative DML · CUB200-2011 (Inception-BN) | Recall@1 | Improved |
| add standard NCE as extra task to discriminative DML (ablation) | discriminative DML · CUB200-2011 (Inception-BN) | Recall@1 | Hurt |
| add DeepCluster, RotNet or Instance Discrimination self-supervision to discriminative DML (ablation) | discriminative DML · CUB200-2011 (Inception-BN) | Recall@1 | No clear effect |
| learn the temperature parameter tau in NCE (ablation) | DaNCE · DML benchmarks | Recall@1 | No clear effect |
Later papers that compared with this method
8 later accepted papers report DIVA as a baseline in their results tables; 8 of them report a better number on most of the results they share with it.
- Learning Time-Aware Causal Representation for Model Generalization in Evolving Domains (ICML 2025 · better on 1 of 1 shared result)
- Evolving Domain Generalization via Latent Structure-Aware Sequential Autoencoder (TPAMI 2023 · better on 2 of 2 shared results)
- MetricFormer: A Unified Perspective of Correlation Exploring in Similarity Learning (NeurIPS 2022 · better on 3 of 3 shared results)
- Towards Improved and Interpretable Deep Metric Learning via Attentive Grouping (TPAMI 2022 · better on 2 of 3 shared results)
- Generalizing to Evolving Domains with Latent Structure-Aware Sequential Autoencoder (ICML 2022 · better on 2 of 2 shared results)
- Domain Invariant Representation Learning with Domain Density Transformations (NeurIPS 2021 · better on 3 of 3 shared results)
- Multi-level Distance Regularization for Deep Metric Learning (AAAI 2021 · better on 1 of 1 shared result)
- Learning Intra-Batch Connections for Deep Metric Learning (ICML 2021 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.91)
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.91)
- Is Fairness Only Metric Deep? Evaluating and Addressing Subgroup Gaps in Deep Metric Learning (ICLR 2022 · similarity 0.88)
- Distance Metric Learning with Joint Representation Diversification (ICML 2020 · similarity 0.88)
- Learning Semantic Proxies from Visual Prompts for Parameter-Efficient Fine-Tuning in Deep Metric Learning (ICLR 2024 · similarity 0.87)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · 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.
