Research map / Label-efficient and robust learning / Recognition and long-tail
Hyperbolic Image Embeddings
Check 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 |
|---|---|---|---|
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet with 4-Conv · MiniImageNet 1-shot 5-way | accuracy: 49.42 ± 0.78% -> 54.43 ± 0.20% | Helped |
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet with 4-Conv · MiniImageNet 5-shot 5-way | accuracy: 68.20 ± 0.66% -> 72.67 ± 0.15% | Helped |
| Use hyperbolic embeddings and hyperbolic prototypes | ProtoNet with ResNet18 · MiniImageNet 1-shot 5-way | accuracy: Hyperbolic ProtoNet: 59.47 ± 0.20%; LEO: 61.76 ± 0.08% | Mixed |
| Use hyperbolic embeddings and hyperbolic prototypes | ProtoNet with ResNet18 · MiniImageNet 5-shot 5-way | accuracy: Hyperbolic ProtoNet: 76.84 ± 0.14%; LEO: 77.59 ± 0.12% | Mixed |
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet with 4-Conv · CUB 1-shot 5-way | accuracy: 51.31 ± 0.91% -> 64.02 ± 0.24% | Helped |
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet with 4-Conv · CUB 5-shot 5-way | accuracy: 70.77 ± 0.69% -> 82.53 ± 0.14% | Helped |
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet · Omniglot 1-shot 5-way | accuracy: 98.2 -> 99.0 | Helped |
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet · Omniglot 5-shot 5-way | accuracy: 99.4 -> 99.4 | No clear effect |
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet · Omniglot 1-shot 20-way | accuracy: 95.8 -> 95.9 | Helped |
| Replace Euclidean prototypes and distances with hyperbolic averaging and distance | ProtoNet · Omniglot 5-shot 20-way | accuracy: 98.6 -> 98.15 | Hurt |
| Map descriptors to the Poincare ball and use hyperbolic multinomial logistic regression | ResNet50 person re-identification classifier · Market-1501, dim 32, sch#2 | Rank-1: 68.0 -> 75.9 | Helped |
| Map descriptors to the Poincare ball and use hyperbolic multinomial logistic regression | ResNet50 person re-identification classifier · Market-1501, dim 32, sch#2 | mAP: 43.4 -> 51.9 | Helped |
Closest papers by meaning
- Hyperbolic Embeddings of Supervised Models (NeurIPS 2024 · similarity 0.88)
- Fully Hyperbolic Convolutional Neural Networks for Computer Vision (ICLR 2024 · similarity 0.87)
- Unsupervised Feature Learning with Emergent Data-Driven Prototypicality (CVPR 2024 · similarity 0.86)
- Hyperbolic Category Discovery (CVPR 2025 · similarity 0.86)
- Learning Structured Representations with Hyperbolic Embeddings (NeurIPS 2024 · similarity 0.85)
- HiHPQ: Hierarchical Hyperbolic Product Quantization for Unsupervised Image Retrieval (AAAI 2024 · 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.