Research map / Image generation / Editing and style transfer
The Euclidean Space is Evil: Hyperbolic Attribute Editing for Few-shot Image Generation
Lingxiao Li, Yi Zhang, Shuhui 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 hyperbolic latent space instead of Euclidean (ablation) | HAE · Animal Faces (unseen, 1-shot) | LPIPS: 0.5129 -> 0.5636 | Improved |
| use hyperbolic latent space instead of Euclidean (ablation) | HAE · Flowers (unseen, 1-shot) | FID: 54.62 -> 50.10 | Improved |
| use hyperbolic latent space instead of Euclidean (ablation) | HAE · Animal Faces (unseen, 1-shot) | FID: 25.27 -> 26.33 | Hurt |
| use hyperbolic latent space instead of Euclidean (ablation) | HAE · VGG Faces (unseen, 1-shot) | LPIPS: 0.5908 -> 0.5919 | No clear effect |
| use HAE instead of SAGE | few-shot image generation · Flowers, Animal Faces, VGGFaces | LPIPS: 0.4392 -> 0.4739 (Flowers) | Improved |
| use HAE instead of SAGE | few-shot image generation · Flowers | FID: 43.52 -> 50.10 | Hurt |
| decrease hyperbolic radius for augmentation from 6.0 to 3.0 (ablation) | HAE data augmentation · Animal Faces classification | accuracy: 60.10 -> 54.14 | Hurt |
| decrease hyperbolic radius (ablation) | HAE · Animal Faces | LPIPS diversity: 0.4520 -> 0.6316 | Improved |
| augment training data with HAE images (radius 6.0) | ResNet-18 classifier · Animal Faces classification | accuracy: 58.67 -> 60.10 | Improved |
| augment training data with HAE images | ResNet-18 classifier · VGGFaces classification | accuracy: 77.99 -> 74.44 | Hurt |
Closest papers by meaning
- HypDAE: Hyperbolic Diffusion Autoencoders for Hierarchical Few-shot Image Generation (ICCV 2025 · similarity 0.90)
- Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation (NeurIPS 2024 · similarity 0.85)
- HyperEditor: Achieving Both Authenticity and Cross-Domain Capability in Image Editing via Hypernetworks (AAAI 2024 · similarity 0.85)
- Unsupervised Feature Learning with Emergent Data-Driven Prototypicality (CVPR 2024 · similarity 0.85)
- DomainGallery: Few-shot Domain-driven Image Generation by Attribute-centric Finetuning (NeurIPS 2024 · similarity 0.84)
- FluxSpace: Disentangled Semantic Editing in Rectified Flow Models (CVPR 2025 · similarity 0.84)
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.
