Research map / Optimization / Optimizers
Smooth-AP: Smoothing the Path Towards Large-Scale Image Retrieval
Andrew Brown, Weidi Xie, Vicky Kalogeiton, Andrew Zisserman
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 |
|---|---|---|---|
| append Smooth-AP loss to ArcFace | ArcFace · VGGFace2 test face retrieval | mAP: 0.858 -> 0.902 | Improved |
| append Smooth-AP loss to ArcFace | ArcFace · IJB-C face retrieval | mAP: 0.772 -> 0.803 | Improved |
| append Smooth-AP loss to SENet-50 softmax model | SENet-50 · VGGFace2 test face retrieval | mAP: 0.828 -> 0.850 | Improved |
| use Smooth-AP instead of triplet loss for fine-tuning | ArcFace · VGGFace2 test face retrieval | mAP: 0.880 -> 0.902 | Improved |
| use Smooth-AP instead of pairwise contrastive loss | SENet-50 · IJB-C face retrieval | mAP: 0.728 -> 0.754 | Improved |
| use Smooth-AP instead of Blackbox AP | ResNet50 · INaturalist retrieval | Recall@1 (%): 62.9 -> 65.9 | Improved |
| increase Smooth-AP batch size from 224 to 384 (ablation) | Smooth-AP · INaturalist retrieval | Recall@1 (%): 65.9 -> 67.2 | Improved |
| lower sigmoid temperature tau from 0.1 to 0.01 (ablation) | Smooth-AP · VGGFace2 test (SENet-50 ablation, mAP) | mAP: 0.824 -> 0.844 | Improved |
| lower sigmoid temperature tau from 0.01 to 0.001 (ablation) | Smooth-AP · VGGFace2 test (SENet-50 ablation, mAP) | mAP: 0.844 -> 0.839 | Hurt |
| increase positive-set size |P| from 4 to 16 (ablation) | Smooth-AP · VGGFace2 test (SENet-50 ablation, mAP) | mAP: 0.844 -> 0.824 | Hurt |
| increase batch size from 64 to 256 (ablation) | Smooth-AP · VGGFace2 test (SENet-50 ablation, mAP) | mAP: 0.824 -> 0.853 | Improved |
| use Smooth-AP instead of FastAP on VehicleID small | ResNet50 · VehicleID (Small test set) | Recall@1 (%): 91.9 -> 94.9 | Improved |
Later papers that compared with this method
2 later accepted papers report SmoothAP as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Exploring the Algorithm-Dependent Generalization of AUPRC Optimization with List Stability (NeurIPS 2022 · better on 9 of 9 shared results)
- Stochastic Optimization of Areas Under Precision-Recall Curves with Provable Convergence (NeurIPS 2021 · better on 5 of 5 shared results)
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
- Robust and Decomposable Average Precision for Image Retrieval (NeurIPS 2021 · similarity 0.89)
- Hierarchical Average Precision Training for Pertinent Image Retrieval (ECCV 2022 · similarity 0.88)
- Learning With Average Precision: Training Image Retrieval With a Listwise Loss (ICCV 2019 · similarity 0.87)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 2024 · similarity 0.87)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.84)
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.83)
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.
