Research map / Vision-language and multimodal / Human and embodied modeling
FreestyleRet: Retrieving Images from Style-Diversified Queries
Hao Li, Yanhao Jia, Peng Jin, Zesen Cheng, Kehan Li, Jialu Sui, Chang Liu, Li Yuan
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 |
|---|---|---|---|
| apply style-init prompt tuning (FreestyleRet) to CLIP encoder | CLIP · DSR Sketch->Image | R@1: 63.6 -> 80.6 | Improved |
| apply FreestyleRet to CLIP for text queries | CLIP · DSR Text->Image | R@1: 72.2 -> 69.9 | Hurt |
| prompt tune cross-modality models for style-diversified inputs | CLIP · DSR Sketch->Image | R@1: 47.5 -> 63.6 | Improved |
| add auxiliary style query (sketch, art, low-res) alongside text query | FreestyleRet · DSR Sketch+Text->Image | R@1: 69.9 -> 82.5 | Improved |
| add auxiliary sketch or art query alongside text query | CLIP* and BLIP* baselines · DSR Art+Text->Image | R@1: 72.2 -> 57.8 (CLIP); 74.3 -> 58.3 (BLIP) | Hurt |
| initialize prompts from style space (shallow) and gram matrix (deep) instead of random (ablation) | FreestyleRet · DSR Sketch->Image | R@1: 78.1 -> 80.6 | Improved |
| insert style-initialized prompts in both shallow and deep layers instead of one (ablation) | FreestyleRet · DSR Low-Res->Image | R@1: 84.7 -> 86.4 | Improved |
| use 4 prompt tokens instead of 1, 2 or 8 (ablation) | FreestyleRet · DSR | R@1 | Improved |
| apply FreestyleRet on a CLIP encoder instead of BLIP on ImageNet-X low-res queries | FreestyleRet · ImageNet-X Low-Res->Image | R@1: 97.5 (BLIP) vs 69.4 (CLIP) | Hurt |
| add style extraction and prompt tuning on frozen encoder | CLIP · Efficiency | trainable parameters and inference time: 427M -> 476M; 68ms -> 96ms | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- LanguageBind — LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment (ICLR 2024) · better on 4 of 4 shared results
- BLIP* — BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (ICML 2022) · better on 1 of 2 shared results
- ImageBind — ImageBind: One Embedding Space To Bind Them All (CVPR 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Hystar: Hypernetwork-driven Style-adaptive Retrieval via Dynamic SVD Modulation (ICLR 2026 · similarity 0.87)
- You'll Never Walk Alone: A Sketch and Text Duet for Fine-Grained Image Retrieval (CVPR 2024 · similarity 0.85)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.84)
- QuRe: Query-Relevant Retrieval through Hard Negative Sampling in Composed Image Retrieval (ICML 2025 · similarity 0.83)
- Highlighting What Matters: Promptable Embeddings for Attribute-Focused Image Retrieval (NeurIPS 2025 · similarity 0.82)
- QuARI: Query Adaptive Retrieval Improvement (NeurIPS 2025 · similarity 0.82)
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.
