Research map / Vision-language and multimodal / Human and embodied modeling
Hierarchical Average Precision Training for Pertinent Image Retrieval
Elias Ramzi, Nicolas Audebert, Nicolas Thome, Clément Rambour, Xavier Bitot
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 smooth upper-bound surrogate loss instead of sigmoid (ablation) | HAPPIER · iNat-base | H-AP: 52.3 -> 53.1 | Improved |
| add clustering loss (ablation) | HAPPIER · iNat-base | H-AP: 53.1 -> 54.3 | Improved |
| train with H-AP instead of fine-level losses | ResNet-50 · SOP | H-AP: +16.1pt on SOP | Improved |
| train with HAPPIER instead of CSL | ResNet-50 · SOP | H-AP: +6pt on H-AP, +8pt on ASI and +2.6pts on NDCG | Improved |
| train with HAPPIER instead of best hierarchical baseline | ResNet-50 · SOP | H-AP: +6.3pt on SOP, +4.2pt on iNat-base | Improved |
| use HAPPIER instead of SigmaNSM on DyML-Vehicle | ResNet-34 · DyML-Vehicle | mAP: 32.0 -> 37.0 | Improved |
| use HAPPIER instead of fine-level TL_SH on SOP coarse level | ResNet-50 · SOP | coarse AP: 14.5 -> 58.4 | Improved |
| use HAPPIER instead of best fine-grained method at fine level | ResNet-50 · SOP | fine-level AP: 62.5 -> 60.4 | Hurt |
| raise alpha (HAPPIER_F) to favour fine level (ablation) | HAPPIER · SOP | fine-level AP: 60.4 -> 62.2 | Improved |
| raise alpha (HAPPIER_F) to favour fine level (ablation) | HAPPIER · SOP | coarse-level AP: 58.4 -> 36.0 | Hurt |
| train with H-AP relevance instead of sum wAP (ablation) | HAPPIER · iNat-base | H-AP: 52.0 -> 53.1 | Improved |
| increase lambda above 0.5 for clustering loss weight (ablation) | HAPPIER · iNat-base | H-AP | Hurt |
Closest papers by meaning
- Optimization of Rank Losses for Image Retrieval (TPAMI 2025 · similarity 0.91)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 2024 · similarity 0.85)
- Hierarchical Retrieval: The Geometry and a Pretrain-Finetune Recipe (NeurIPS 2025 · similarity 0.84)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.84)
- QuRe: Query-Relevant Retrieval through Hard Negative Sampling in Composed Image Retrieval (ICML 2025 · similarity 0.84)
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (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.
