Research map / Label-efficient and robust learning / Knowledge distillation
CLIDD: Cross-Layer Independent Deformable Description for Efficient and Discriminative Local Feature Representation
Haodi Yao, Fenghua He, Ning Hao, Yao Su
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 |
|---|---|---|---|
| add cross-layer predictor instead of single-layer offset prediction (ablation) | CLIDD U128 · MegaDepth-1500 | AUC@10: 75.51 -> 77.15 | Improved |
| add cross-layer predictor instead of single-layer offset prediction (ablation) | CLIDD A48 · MegaDepth-1500 | AUC@10: 56.20 -> 57.13 | Improved |
| add layer-independent sampler instead of shared sampler (ablation) | CLIDD U128 · MegaDepth-1500 | AUC@10: 75.80 -> 77.15 | Improved |
| add layer-independent sampler instead of shared sampler (ablation) | CLIDD A48 · MegaDepth-1500 | AUC@10: 55.22 -> 57.13 | Improved |
| use CLIDD instead of Full-Scale SDDH head (ablation) | U128 · MegaDepth-1500 | AUC@10: 75.63 -> 77.15 | Improved |
| use CLIDD instead of Sub-Scale Vanilla head (ablation) | A48 · MegaDepth-1500 | AUC@10: 36.39 -> 57.13 | Improved |
| use CLIDD instead of Full-Scale Vanilla head (ablation) | U128 · MegaDepth-1500 | AUC@10: 67.89 -> 77.15 | Improved |
| add Procrustes distillation (w_OP=1) to DualSoftmax (ablation) | A48 · MegaDepth-1500 | AUC@10: 46.65 -> 57.13 | Improved |
| add DualSoftmax loss (w_DS=0.05) to pure distillation (ablation) | A48 · MegaDepth-1500 | AUC@10: 55.93 -> 57.13 | Improved |
| add distillation to direct loss for larger model (ablation) | T64 · MegaDepth-1500 | AUC@10: 66.44 -> 66.74 | No clear effect |
| replace ALIKED-N32 teacher with U128 teacher (ablation) | A48 · MegaDepth-1500 | AUC@10: 57.13 -> 58.53 | No clear effect |
| replace SDDH with CLIDD on frozen ALIKED-N32 (ablation) | ALIKED-N32 · MegaDepth-1500 | AUC@10: 73.56 -> 74.40 | Improved |
Closest papers by meaning
- RDD: Robust Feature Detector and Descriptor using Deformable Transformer (CVPR 2025 · similarity 0.84)
- RDD: Improving Structure from Motion with a Robust Feature Detector and Descriptor (IJCV 2026 · similarity 0.84)
- Enhancing Deformable Local Features by Jointly Learning To Detect and Describe Keypoints (CVPR 2023 · similarity 0.83)
- Cross-Descriptor Visual Localization and Mapping (ICCV 2021 · similarity 0.83)
- Extracting Deformation-Aware Local Features by Learning to Deform (NeurIPS 2021 · similarity 0.83)
- CroCoDL: Cross-device Collaborative Dataset for Localization (CVPR 2025 · similarity 0.82)
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.
