Research map / Detection and segmentation / CNN features
LIFT: Learned Invariant Feature Transform
Kwang Moo Yi, Eduard Trulls, Vincent Lepetit, Pascal Fua
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 |
|---|---|---|---|
| fully train the detector after pre-training (ablation) | LIFT · Strecha (LIFT trained on Piccadilly) | matching score: .367 -> .374 | Improved |
| fully train the detector after pre-training (ablation) | LIFT · Strecha (LIFT trained on Piccadilly) | repeatability: .436 -> .446 | Improved |
| fully train the detector after pre-training (ablation) | LIFT · Strecha (LIFT trained on Roman-Forum) | matching score: .368 -> .369 | No clear effect |
| replace SIFT orientation with learned orientation (ablation) | SIFT detector and descriptor · Strecha (LIFT trained on Piccadilly) | matching score: .282 -> .341 | Improved |
| replace SIFT descriptor with learned descriptor (ablation) | SIFT detector and orientation · Strecha (LIFT trained on Piccadilly) | matching score: .282 -> .290 | Improved |
| replace SIFT detector with learned detector (ablation) | SIFT orientation and descriptor · Strecha (LIFT trained on Piccadilly) | matching score: .282 -> .325 | Improved |
| use learned detector instead of SIFT detector in full pipeline (ablation) | LIFT with learned orientation and descriptor · Strecha (LIFT trained on Piccadilly) | matching score: .344 -> .374 | Improved |
| use learned orientation instead of SIFT orientation in full pipeline (ablation) | LIFT with learned detector and descriptor · Strecha (LIFT trained on Piccadilly) | matching score: .339 -> .374 | Improved |
| use learned descriptor instead of SIFT descriptor in full pipeline (ablation) | LIFT with learned detector and orientation · Strecha (LIFT trained on Piccadilly) | matching score: .372 -> .374 | Improved |
| use learned detector instead of SIFT detector (ablation) | SIFT orientation and descriptor · Strecha (LIFT trained on Piccadilly) | repeatability: .428 -> .446 | Improved |
| replace SIFT with the full LIFT pipeline | SIFT · Strecha | matching score: .283 -> .374 | Improved |
| replace SIFT with the full LIFT pipeline | SIFT · DTU | matching score: .272 -> .317 | Improved |
Closest papers by meaning
- LiFT: A Surprisingly Simple Lightweight Feature Transform for Dense ViT Descriptors (ECCV 2024 · similarity 0.86)
- RIConv++: Effective Rotation Invariant Convolutions for 3D Point Clouds Deep Learning (IJCV 2022 · similarity 0.84)
- Deep Positional and Relational Feature Learning for Rotation-Invariant Point Cloud Analysis (ECCV 2020 · similarity 0.84)
- LF-Net: Learning Local Features from Images (NeurIPS 2018 · similarity 0.84)
- Revisiting Transformation Invariant Geometric Deep Learning: An Initial Representation Perspective (TPAMI 2025 · similarity 0.84)
- 3D-LFM: Lifting Foundation Model (CVPR 2024 · 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.
