Research map / Architectures and efficiency / Pruning and efficient tuning
LoRAverse: A Submodular Framework to Retrieve Diverse Adapters for Diffusion Models
Mert Sonmezer, Matthew Zheng, Pinar Yanardag
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 LoRAverse submodular retrieval instead of Stylus | Stable Diffusion with LoRA retrieval · MS-COCO + PartiPrompts (CFG=7) | TCE: 20.30 -> 22.63 | Improved |
| use LoRAverse submodular retrieval instead of Stylus | Stable Diffusion with LoRA retrieval · MS-COCO + PartiPrompts (CFG=7) | TIE: 38.53 -> 40.06 | Improved |
| use LoRAverse submodular retrieval instead of Stylus | Stable Diffusion with LoRA retrieval · MS-COCO + PartiPrompts (CFG=7) | I2I: 0.825 -> 0.784 | Improved |
| use LoRAverse submodular retrieval instead of Stylus | Stable Diffusion with LoRA retrieval · MS-COCO + PartiPrompts (CFG=7) | CLIP: 25.41 -> 25.07 | Hurt |
| use LoRAverse instead of Stylus | Stable Diffusion with LoRA retrieval · User study | US-D: 2.82 -> 3.92 | Improved |
| use LoRAverse instead of Stylus | Stable Diffusion with LoRA retrieval · User study | US-F: 3.71 -> 3.78 | No clear effect |
| use LoRAverse instead of Stylus | Stable Diffusion with LoRA retrieval · User study | US-P: 29.90% -> 47.55% | Improved |
| use submodular retrieval instead of cosine similarity retrieval (ablation) | LoRAverse · Retrieval ablation (520 images) | TCE: 23.47 -> 23.97 | Improved |
| use submodular retrieval instead of cosine similarity retrieval (ablation) | LoRAverse · Retrieval ablation (520 images) | I2I: 0.781 -> 0.762 | Improved |
| use submodular retrieval instead of cosine similarity retrieval (ablation) | LoRAverse · Retrieval ablation (520 images) | CLIP: 24.67 -> 24.50 | Hurt |
| use submodular retrieval instead of RTop-100 random sampling (ablation) | LoRAverse · Additional baselines (K=8) | TIE: 42.40 -> 43.04 | Improved |
| use submodular retrieval instead of RTop-100 random sampling (ablation) | LoRAverse · Additional baselines (K=8) | CLIP: 24.99 -> 24.94 | No clear effect |
Closest papers by meaning
- LoRAtorio: An intrinsic approach to LoRA Skill Composition (NeurIPS 2026 · similarity 0.91)
- LoRA-X: Bridging Foundation Models with Training-Free Cross-Model Adaptation (ICLR 2025 · similarity 0.89)
- LoRACLR: Contrastive Adaptation for Customization of Diffusion Models (CVPR 2025 · similarity 0.89)
- Orthogonal Adaptation for Modular Customization of Diffusion Models (CVPR 2024 · similarity 0.88)
- UnlearnCanvas: Stylized Image Dataset for Enhanced Machine Unlearning Evaluation in Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Time-Varying LoRA: Towards Effective Cross-Domain Fine-Tuning of Diffusion Models (NeurIPS 2024 · 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.
