Research map / Detection and segmentation / CNN features
Learning Discriminative and Transformation Covariant Local Feature Detectors
Xu Zhang, Felix X. Yu, Svebor Karaman, Shih-Fu Chang
Check 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 the proposed transformation predictor with standard patches and identity loss instead of T-P24 | Local feature detection · Webcam, 1000 features | repeatability (%): 61.7 -> 68.4 | Improved |
| Use the proposed transformation predictor with standard patches and identity loss instead of T-P24 | Local feature detection · Webcam, 200 features | repeatability (%): 45.1 -> 52.6 | Improved |
| Use the proposed transformation predictor with standard patches and identity loss instead of T-P24 | Local feature detection · EF, 1000 features | repeatability (%): 45.4 -> 46.6 | Improved |
| Use the proposed transformation predictor with standard patches and identity loss instead of HarLap | Local feature detection · EF, 200 features | repeatability (%): 33.4 -> 36.3 | Improved |
| Use the proposed transformation predictor with standard patches and identity loss instead of HesLap | Local feature detection · VGG-Affine, 1000 features | repeatability (%): 66.7 -> 70.2 | Improved |
| Use the proposed transformation predictor with standard patches and identity loss instead of HesLap | Local feature detection · VGG-Affine, 200 features | repeatability (%): 60.0 -> 61.2 | Improved |
| Use T-P24-detected patches as standard patches rather than randomly sampled patches (ablation) | Proposed detector · Repeatability across VGG-Affine, EF, and Webcam | repeatability (%) | Improved |
| Use T-P24-detected patches as standard patches rather than SIFT-detected patches (ablation) | Proposed detector · Repeatability across VGG-Affine, EF, and Webcam | repeatability (%) | Improved |
| Use T-P24-detected patches as standard patches rather than Hessian-Affine-detected patches (ablation) | Proposed detector · Repeatability across VGG-Affine, EF, and Webcam | repeatability (%) | Improved |
| Add the identity loss with an adequate nonzero trade-off rather than using only the covariant constraint (ablation) | Proposed detector · Repeatability across VGG-Affine, EF, and Webcam | repeatability (%) | Improved |
| Use α=0, removing the identity-loss contribution (ablation) | Proposed detector · Repeatability across VGG-Affine, EF, and Webcam | repeatability (%) | Hurt |
| Use α=100, relying on the identity constraint rather than balancing it with covariance (ablation) | Proposed detector · Repeatability across VGG-Affine, EF, and Webcam | repeatability (%) | Hurt |
Closest papers by meaning
- Online Invariance Selection for Local Feature Descriptors (ECCV 2020 · similarity 0.85)
- Co-Attentive Equivariant Neural Networks: Focusing Equivariance On Transformations Co-Occurring in Data (ICLR 2020 · similarity 0.85)
- Repeatability Is Not Enough: Learning Affine Regions via Discriminability (ECCV 2018 · similarity 0.85)
- Self-Supervised Learning Based on Transformed Image Reconstruction for Equivariance-Coherent Feature Representation (AAAI 2026 · similarity 0.84)
- Learning an Invariant and Equivariant Network for Weakly Supervised Object Detection (TPAMI 2023 · similarity 0.84)
- Self-supervised Transformation Learning for Equivariant Representations (NeurIPS 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.
