Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learning Rotation-Equivariant Features for Visual Correspondence
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 group-aligning instead of group pooling (ablation) | Rotation-equivariant descriptors · Roto-360 | MMA @3px: 57.92 -> 90.18 | Helped |
| Add orientation alignment loss (ablation) | Proposed descriptor method · Roto-360 | MMA @3px: 83.26 -> 90.18 | Helped |
| Add contrastive descriptor loss (ablation) | Proposed descriptor method · Roto-360 | MMA @3px: 24.98 -> 90.18 | Helped |
| Use an image pyramid at inference (ablation) | Proposed descriptor method · HPatches all | MMA @3px: 62.25 -> 63.42 | Helped |
| Use an image pyramid at inference (ablation) | Proposed descriptor method · Roto-360 | MMA @3px: 90.43 -> 90.18 | No clear effect |
| Use a rotation-equivariant backbone rather than conventional ResNet-18 (ablation) | Descriptor feature extractor · Roto-360 | MMA @3px: 8.51 -> 90.18 | Helped |
| Use multiple orientation-candidate descriptors (ablation) | Group-aligning descriptors · Roto-360 | MMA @3px: 90.18 -> 91.69 | Helped |
| Use ReWRN instead of ReResNet-18 (ablation) | Group-aligning descriptor method · HPatches all | MMA @3px: 63.42 -> 69.70 | Helped |
| Set cyclic group order to 16 rather than 64 (ablation) | Equivariant descriptor network · Roto-360 | MMA @3px: 83.32 -> 90.18 | Helped |
| Set cyclic group order to 16 rather than 36 (ablation) | Equivariant descriptor network · Roto-360 | MMA @3px: 85.89 -> 90.18 | Helped |
| Set cyclic group order to 16 rather than 32 (ablation) | Equivariant descriptor network · Roto-360 | MMA @3px: 87.31 -> 90.18 | Helped |
| Set cyclic group order to 16 rather than 24 (ablation) | Equivariant descriptor network · Roto-360 | MMA @3px: 88.34 -> 90.18 | Helped |
Closest papers by meaning
- RotDCF: Decomposition of Convolutional Filters for Rotation-Equivariant Deep Networks (ICLR 2019 · similarity 0.90)
- Relaxed Rotational Equivariance via G-Biases in Vision (AAAI 2025 · similarity 0.89)
- GIFT: Learning Transformation-Invariant Dense Visual Descriptors via Group CNNs (NeurIPS 2019 · similarity 0.88)
- Variational Partial Group Convolutions for Input-Aware Partial Equivariance of Rotations and Color-Shifts (ICML 2024 · similarity 0.88)
- R2Det: Exploring Relaxed Rotation Equivariance in 2D Object Detection (ICLR 2025 · similarity 0.87)
- Rotation-Equivariant Self-Supervised Method in Image Denoising (CVPR 2025 · similarity 0.87)
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.