Research map / Label-efficient and robust learning / Recognition and long-tail
Rethinking Rotation-Invariant Recognition of Fine-Grained Shapes From the Perspective of Contour Points
Yanjie Xu, Yunzhi Chen, Handing Xu, Tianmu Wang, Yaguan Li, Zhenguo Nie
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 LOA instead of LRA (ablation) | contour rotation-invariant network · Kidney retrieval, lambda=1 | retrieval success rate: 6.65 -> 90.78 | Improved |
| use LOA instead of LRA (ablation) | contour rotation-invariant network · Kimia99 retrieval, lambda=1 | retrieval success rate: 18.39 -> 76.01 | Improved |
| use LOA instead of LRA with more augmentation (ablation) | contour rotation-invariant network · Kimia99 retrieval, lambda=5 | retrieval success rate: 98.29 -> 96.45 | Hurt |
| use LOA instead of LRA (ablation) | contour rotation-invariant network · Kidney retrieval, lambda=5 | retrieval success rate: 80.09 -> 99.33 | Improved |
| increase rotation augmentation lambda from 1 to 3 | Ours · Flavia classification | mean accuracy: 89.14 -> 95.96 | Improved |
| increase rotation augmentation lambda from 1 to 5 | Ours · Kidney retrieval | mean retrieval accuracy: 94.20 -> 100.00 | Improved |
| increase rotation augmentation lambda from 1 to 5 | Ours · Liver retrieval | mean retrieval accuracy: 65.07 -> 99.49 | Improved |
| add noisy shapes as training augmentation (Ours_DA) | Ours · contour noise levels 0-5 retrieval | recognition accuracy | Improved |
| use contour-point encoding instead of pixel-based networks | Ours vs image-based methods · contour noise robustness | recognition accuracy | Hurt |
| compare contour-point method against RIC-CNN on Flavia classification | Ours · Flavia classification | accuracy (%): 88.68 -> 97.11 | Improved |
| rotation augmentation of training data on rotation-center robustness | Ours · rotation center position | accuracy | Mixed |
| change point-sequence starting point (Type-II rotation) | Ours · rotation type test | accuracy | No clear effect |
Closest papers by meaning
- PaRot: Patch-Wise Rotation-Invariant Network via Feature Disentanglement and Pose Restoration (AAAI 2023 · similarity 0.86)
- Learning Canonical 3D Object Representation for Fine-Grained Recognition (ICCV 2021 · similarity 0.86)
- Rethinking Rotation Invariance with Point Cloud Registration (AAAI 2023 · similarity 0.86)
- ShapeCodes: Self-Supervised Feature Learning by Lifting Views to Viewgrids (ECCV 2018 · similarity 0.86)
- Selective Disk Bispectrum: A Complete and Rotation Invariant Image Descriptor (NeurIPS 2026 · similarity 0.85)
- ShapeEmbed: a self-supervised learning framework for 2D contour quantification (NeurIPS 2025 · similarity 0.85)
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.
