Research map / Image generation / Text-to-image diffusion
ChimeraLoRA: Multi-Head LoRA-Guided Synthetic Datasets
Hoyoung Kim, Minwoo Jang, Jabin Koo, Sangdoo Yun, Jungseul Ok
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 |
|---|---|---|---|
| add multi-head LoRA (shared A, per-image B) to per-image LoRA (ablation) | LoFT-style baseline · FGVCAircraft 4-shot | accuracy (%): 41.7 -> 43.9 | Improved |
| add semantic boosting (bounding-box-preserving crop) to per-image LoRA (ablation) | LoFT-style baseline · FGVCAircraft 4-shot | accuracy (%): 41.7 -> 44.4 | Improved |
| combine multi-head LoRA and semantic boosting (ablation) | ChimeraLoRA · FGVCAircraft 4-shot | accuracy (%): 41.7 -> 46.0 | Improved |
| add semantic boosting on top of multi-head LoRA (ablation) | ChimeraLoRA · Flowers102 4-shot | accuracy (%): 93.1 -> 93.4 | Improved |
| add semantic boosting without multi-head LoRA (ablation) | LoFT-style baseline · Flowers102 4-shot | accuracy (%): 91.3 -> 92.2 | Improved |
| share LoRA B instead of LoRA A across images (ablation) | ChimeraLoRA · qualitative (headphone, motorbike, stapler, panda) | image fidelity | Hurt |
| add 500 synthetic images per class with ChimeraLoRA to 4-shot real data | CLIP ViT-B/16 fine-tuning · 4-shot, 9 datasets (average) | accuracy (%): 71.8 -> 74.6 | Improved |
| add 500 synthetic images per class with DataDream | CLIP ViT-B/16 fine-tuning · 4-shot, 9 datasets (average) | accuracy (%): 71.8 -> 71.3 | Hurt |
| add synthetic images to tail classes with ChimeraLoRA | CLIP ViT-B/16 long-tail training · EuroSAT long-tail (tail classes) | accuracy (%): 13.6 -> 51.3 | Improved |
| increase synthetic images per class (ablation) | DataDream and LoFT · EuroSAT, FGVCAircraft | accuracy (%) | Hurt |
| increase synthetic images per class (ablation) | ChimeraLoRA · EuroSAT, FGVCAircraft | accuracy (%) | Improved |
| sample Dirichlet mixture weights per image vs uniform weights 1/K (ablation) | ChimeraLoRA · synthesis | coverage | Improved |
Closest papers by meaning
- LoRAtorio: An intrinsic approach to LoRA Skill Composition (NeurIPS 2026 · similarity 0.90)
- LoRACLR: Contrastive Adaptation for Customization of Diffusion Models (CVPR 2025 · similarity 0.88)
- LoRA-X: Bridging Foundation Models with Training-Free Cross-Model Adaptation (ICLR 2025 · similarity 0.88)
- LoRA Recycle: Unlocking Tuning-Free Few-Shot Adaptability in Visual Foundation Models by Recycling Pre-Tuned LoRAs (CVPR 2025 · similarity 0.87)
- Cached Multi-Lora Composition for Multi-Concept Image Generation (ICLR 2025 · similarity 0.87)
- CORAL: Disentangling Latent Representations in Long-Tailed Diffusion (NeurIPS 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.
