Research map / Image generation / Autoregressive and latent generation
Deep Image Retrieval: Learning Global Representations for Image Search
Albert Gordo, Jon Almazán, Jerome Revaud, Diane Larlus
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 |
|---|---|---|---|
| fine-tune with ranking loss (R-Clean) instead of classification (C-Clean) (ablation) | R-MAC VGG16 rigid grid · Oxford 5k | mAP: 75.2 -> 81.1 | Improved |
| fine-tune with ranking loss (R-Clean) instead of classification (C-Clean) (ablation) | R-MAC VGG16 rigid grid · Paris 6k | mAP: 83.2 -> 86.0 | Improved |
| fine-tune for classification on Landmarks (ablation) | R-MAC VGG16 · Oxford 5k | mAP: 66.2 -> 74.8 | Improved |
| clean the Landmarks dataset for classification fine-tuning (ablation) | C-Full · Oxford 5k | mAP: 74.8 -> 75.2 | No clear effect |
| clean the Landmarks dataset for classification fine-tuning (ablation) | C-Full · Paris 6k | mAP: 82.5 -> 83.2 | No clear effect |
| train ranking network on clean data (triplet loss intolerant to outliers) (ablation) | R-MAC ranking · Oxford 5k | mAP | Improved |
| initialize ranking training from a C-Full classification model (ablation) | R-MAC ranking · Oxford 5k | mAP | Improved |
| learn PCA on Landmarks instead of Paris/Oxford (ablation) | R-MAC reimplementation · Oxford 5k | mAP: 66.9 -> 66.2 | Hurt |
| replace rigid grid with 256 RPN proposals (ablation) | R-Clean · Oxford 5k | mAP: 81.1 -> 83.1 | Improved |
| replace rigid grid with 256 RPN proposals (ablation) | R-Clean · Paris 6k | mAP: 86.0 -> 87.1 | Improved |
| replace rigid grid with 256 RPN proposals (ablation) | C-Full · Oxford 5k | mAP: 74.8 -> 76.8 | Improved |
| use only 16 proposals instead of rigid grid (ablation) | R-Clean · Paris 6k | mAP: 86.0 -> 85.4 | Hurt |
Closest papers by meaning
- End-to-End Learning of Deep Visual Representations for Image Retrieval (IJCV 2017 · similarity 0.95)
- SIFT Meets CNN: A Decade Survey of Instance Retrieval (TPAMI 2017 · similarity 0.91)
- Unifying Deep Local and Global Features for Image Search (ECCV 2020 · similarity 0.91)
- Learning Super-Features for Image Retrieval (ICLR 2022 · similarity 0.91)
- Learning Spatial-context-aware Global Visual Feature Representation for Instance Image Retrieval (ICCV 2023 · similarity 0.91)
- Large-Scale Image Retrieval With Attentive Deep Local Features (ICCV 2017 · similarity 0.90)
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.
