Research map / Detection and segmentation / CNN features
SIFT Matching by Context Exposed
Fabio Bellavia
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 many-to-many matching (f'=5) instead of one-to-one (f'=1) in blob matching (ablation) | Blob matching HarrisZ+SOSNet · Planar and non-planar datasets | mAP (%): around 57 -> 68.8 | Improved |
| restrict rank pre-filter to the top 10 matches (10_union) instead of all (Omega) (ablation) | Blob matching · Planar and non-planar datasets | mAP (%): 68.8 vs 68.7 | No clear effect |
| use HarrisZ keypoints instead of SIFT (ablation) | Blob matching · Planar and non-planar datasets | mAP | Improved |
| add second stage (contraction/expansion) to DTM (ablation) | DTM · Global statistics | recall (%): 67.84 -> 78.85 | Improved |
| add second stage to DTM (ablation) | DTM · Global statistics | precision (%): 78.78 -> 79.82 | No clear effect |
| use D+ similarity instead of D_geq for NNR (ablation) | Blob matching with f'=5 · Planar and non-planar datasets | mAP | Improved |
| add FGINN (t_o=75% or 10 px) on top of D+ (ablation) | Blob matching · Planar and non-planar datasets | mAP | Improved |
| add FGINN with D_geq^+ instead of D_geq (ablation) | Blob matching · Planar and non-planar datasets | mAP | No clear effect |
| combine both matching directions with harmonic mean instead of one reference image (ablation) | Blob matching · Planar and non-planar datasets | mAP | Improved |
| add 1SAC post-filtering (ablation) | Local spatial filters · Planar, non-planar, SUN3D | precision/recall | Mixed |
| train ACNe on SUN3D instead of YFCC100M (ablation) | ACNe · SUN3D | precision/recall | Improved |
| switch from baseline (SIFT+RootSIFT, one-to-one) to best config (HarrisZ+SOSNet, many-to-many) (ablation) | Matching pipeline with 1SAC · Planar and non-planar datasets | correct matches: correct matches roughly doubled, output matches five times | Improved |
Closest papers by meaning
- SIFTing Through Scales (TPAMI 2016 · similarity 0.85)
- Comparative Evaluation of Hand-Crafted and Learned Local Features (CVPR 2017 · similarity 0.85)
- SGAD: Semantic and Geometric-aware Descriptor for Local Feature Matching (ICCV 2025 · similarity 0.84)
- Dense Correspondences across Scenes and Scales (TPAMI 2016 · similarity 0.84)
- Raising the Ceiling: Conflict-Free Local Feature Matching with Dynamic View Switching (ECCV 2024 · similarity 0.84)
- MATCHA: Towards Matching Anything (CVPR 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.
