Research map / Label-efficient and robust learning / Recognition and long-tail
Embedding Based on Function Approximation for Large Scale Image Search
Thanh-Toan Do, Ngai Man Cheung
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 function-approximation embedding (FAemb) instead of VLAT at same dimension | embedding at n=8 · Holidays and Oxford5k | mAP: +1.8% and +3.9% | Improved |
| use FAemb instead of VLAT at n=16, 32 | embedding · Holidays and Oxford5k | mAP: about +3% | Improved |
| use FAemb (n=16, D=15,525) instead of VLAD/Temb (n=128, D=16,256) | embedding · Holidays | mAP: +7.5%/+2% | Improved |
| use FAemb (n=16, D=15,525) instead of VLAD/Temb (n=128, D=16,256) | embedding · Oxford5k | mAP: +8.1%/+5% | Improved |
| use fast closed-form F-FAemb instead of FAemb | embedding · Holidays/Oxford5k, n=16 | mAP: 75.8 -> 75.5 (Holidays); 67.7 -> 67.6 (Oxford5k) | No clear effect |
| add rotation normalization (RN) to F-FAemb (n=32) | F-FAemb · Holidays and Oxford5k | mAP: 77.0 -> 80.7; 70.7 -> 74.2 | Improved |
| increase number of anchor points n | embedding methods · Holidays and Oxford5k | mAP | Improved |
| use F-FAemb +RN instead of Temb+RN at D=7,245/8,064 | embedding · Holidays | mAP: 77.1 -> 75.5 | Hurt |
| use F-FAemb +RN instead of CNN features | image representation · Oxford5k | mAP: 54.5 -> 74.2 | Improved |
| use F-FAemb +RN instead of Temb+RN at short dimensions | short representation · Holidays+Flickr1M | mAP: 49.4 -> 68.5 at D=1,024 | Improved |
| use F-FAemb instead of Temb as binary codes (ITQ) | binary codes · Holidays | mAP: 57.3 -> 59.7 at 1024 bits | Improved |
| use F-FAemb instead of Temb at 128 bits | binary codes · Oxford5k | mAP: 27.1 -> 26.4 | Hurt |
Closest papers by meaning
- Collaborative Index Embedding for Image Retrieval (TPAMI 2017 · similarity 0.87)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.86)
- Efficient Large-Scale Similarity Search Using Matrix Factorization (CVPR 2016 · similarity 0.86)
- Local Orthogonal-Group Testing (ECCV 2018 · similarity 0.85)
- Revisiting Self-Similarity: Structural Embedding for Image Retrieval (CVPR 2023 · similarity 0.85)
- QuARI: Query Adaptive Retrieval Improvement (NeurIPS 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.
