Research map / Image generation / Text-to-image diffusion
CARLoS: Retrieval via Concise Assessment Representation of LoRAs at Scale
Shahar Sarfaty, Adi Haviv, Uri Hacohen, Niva Elkin-Koren, Roi Livni, Amit H. Bermano
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 Strength filtering (ablation) | CARLoS retrieval · LoRA retrieval top-3 | SigLIP2: 0.335 -> 0.350 | Improved |
| add Strength filtering (ablation) | CARLoS retrieval · LoRA retrieval top-3 | ImageReward: 0.495 -> 0.505 | Improved |
| add Strength filtering (ablation) | CARLoS retrieval · LoRA retrieval top-3 | HPS: 0.596 -> 0.596 | No clear effect |
| add Consistency filtering (ablation) | CARLoS retrieval · LoRA retrieval top-3 | SigLIP2: 0.342 -> 0.350 | Improved |
| add Consistency filtering (ablation) | CARLoS retrieval · LoRA retrieval top-3 | HPS: 0.599 -> 0.596 | Hurt |
| use query as suffix instead of prefix (ablation) | CARLoS retrieval · LoRA retrieval top-3 | ImageReward: 0.488 -> 0.505 | Improved |
| use suffix instead of prefix and suffix (ablation) | CARLoS retrieval · LoRA retrieval top-3 | ImageReward: 0.495 -> 0.505 | Improved |
| use reciprocal CLIP-diff with base prompts instead of only the query (ablation) | CARLoS retrieval · LoRA retrieval top-3 | ImageReward: 0.426 -> 0.505 | Improved |
| use reciprocal CLIP-diff with base prompts instead of only the query (ablation) | CARLoS retrieval · LoRA retrieval top-3 | HPS: 0.538 -> 0.596 | Improved |
| use generation-based CARLoS instead of text-metadata retrieval (Qwen3) | CARLoS retrieval · LoRA retrieval top-3 | SigLIP2: 0.307 -> 0.350 | Improved |
| use generation-based CARLoS instead of text-metadata retrieval (E5) | CARLoS retrieval · Retrieval diversity top-3 | Effective LoRA count: 157.069 -> 237.598 | Improved |
| use CARLoS instead of BGE text retrieval | CARLoS retrieval · LoRA retrieval top-3 | SigLIP2: 0.199 -> 0.350 | Improved |
Closest papers by meaning
- Merging LoRAs like Playing LEGO: Pushing the Modularity of LoRA to Extremes Through Rank-Wise Clustering (ICLR 2025 · similarity 0.85)
- INQUIRE: A Natural World Text-to-Image Retrieval Benchmark (NeurIPS 2024 · similarity 0.84)
- SANEval: Open-Vocabulary Compositional Benchmarks with Failure-mode Diagnosis (NeurIPS 2026 · similarity 0.84)
- GenIR: Generative Visual Feedback for Mental Image Retrieval (NeurIPS 2025 · similarity 0.84)
- Revisiting text-to-image evaluation with Gecko: on metrics, prompts, and human rating (ICLR 2025 · similarity 0.84)
- TIGeR: Unifying Text-to-Image Generation and Retrieval with Large Multimodal Models (ICLR 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.
