Research map / Detection and segmentation / CNN features
Hyperbolic Hierarchical Clustering for Visual Representation Learning
Jianan Wei, Guikun Chen, Zhiyuan Weng, Chunchao Guo, Yujia Wang, Wenguan Wang
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 hierarchical clustering instead of shifted windows (ablation) | ClusterMixer · ImageNet-1K val | top-1: 72.7 -> 73.4 | Improved |
| use hyperbolic similarity for window-level clustering instead of Euclidean (ablation) | HCFormer-Tiny · ImageNet-1K val | top-1: 73.4 -> 75.1 | Improved |
| use hyperbolic similarity at patch level too (ablation) | HCFormer-Tiny · ImageNet-1K val | top-1: 75.1 -> 74.7 | Hurt |
| vary curvature kappa 0.1 / 1.0 / 10.0 (ablation) | HCFormer-Tiny · ImageNet-1K val | top-1: 74.9 / 75.1 / 74.8 | No clear effect |
| more window-level clusters (9 -> 16 -> 25) (ablation) | HCFormer-Tiny · ImageNet-1K val | top-1: 74.9 -> 75.1 -> 74.6 | Mixed |
| add hierarchical clustering (ablation) | HCFormer-Tiny · ImageNet-1K val | top-1: 72.7 -> 75.1 | Improved |
| add hyperbolic geometry (ablation) | HCFormer-Tiny · ImageNet-1K val | top-1: 73.4 -> 75.1 | Improved |
| add relative positional bias (ablation) | HCFormer-Tiny · ImageNet-1K val | top-1: 74.4 -> 75.1 | Improved |
| add hierarchical clustering (ablation) | HCFormer-Tiny · ADE20K val | mIoU: 34.6 -> 37.3 | Improved |
| add hyperbolic geometry (ablation) | HCFormer-Tiny · ADE20K val | mIoU: 35.2 -> 37.3 | Improved |
| add hierarchical clustering + hyperbolic geometry + rel. pos. to ClusterMixer baseline (ablation) | HCFormer-Tiny · COCO val2017 | AP box: 34.3 -> 36.4 | Improved |
| HCFormer vs prior clustering backbones at similar size | HCFormer · ImageNet-1K val | top-1: FEC-Small 72.7 -> HCFormer-Nano 73.0 | Improved |
Closest papers by meaning
- Semantic Equitable Clustering: A Simple and Effective Strategy for Clustering Vision Tokens (ICCV 2025 · similarity 0.89)
- ClusterFomer: Clustering As A Universal Visual Learner (NeurIPS 2023 · similarity 0.88)
- Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning (CVPR 2026 · similarity 0.87)
- HyperVAttention: Efficient Sparse Attention with Spatio-Temporal Clustering for Video Diffusion (NeurIPS 2026 · similarity 0.86)
- Revisiting the Integration of Convolution and Attention for Vision Backbone (NeurIPS 2024 · similarity 0.86)
- Improving Autoregressive Visual Generation with Cluster-Oriented Token Prediction (CVPR 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.
