Research map / Vision-language and multimodal / Human and embodied modeling
Beyond Semantic Search: Towards Referential Anchoring in Composed Image Retrieval
Yuxin Yang, Yinan Zhou, Yuxin Chen, Ziqi Zhang, Zongyang Ma, Chunfeng Yuan, Bing Li, Jun Gao, Weiming Hu
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 CAAM (adaptive attention modulation) to baseline (ablation) | AdaFocal · OACIRR | Avg recall: 74.91 -> 79.00 | Improved |
| use Transformer CRM instead of average pooling (ablation) | CAAM · OACIRR | Avg recall: 76.39 -> 78.19 (frozen probes) | Improved |
| make probe tokens learnable (Transformer CRM) (ablation) | CAAM · OACIRR | Avg recall: 78.19 -> 79.00 | Improved |
| make probe tokens learnable (average pooling) (ablation) | CAAM · OACIRR | Avg recall: 76.39 -> 76.31 | No clear effect |
| fix a positive modulation scalar beta instead of beta=0 (ablation) | AdaFocal · OACIRR | RID@1 / R@1 | Improved |
| use too large fixed beta (ablation) | AdaFocal · OACIRR | R@1 | Hurt |
| predict beta per query (CAAM) instead of fixed beta (ablation) | AdaFocal · OACIRR | RID@1 / R@1 | Improved |
| train SPRC on OACIRR instead of CIRR (ablation) | SPRC (ViT-G) · OACIRR | Avg recall: 37.30 -> 74.05 | Improved |
| use AdaFocal instead of OACIRR-trained SPRC (ViT-G) | AdaFocal · OACIRR | RID@1: avg +7.47 RID@1, +4.14 R@1 | Improved |
| scale backbone from ViT-L to ViT-G | AdaFocal · OACIRR | Avg recall: 76.40 -> 79.00 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- U-MARVEL — U-MARVEL: Unveiling Key Factors for Universal Multimodal Retrieval via Embedding Learning with MLLMs (ICLR 2026) · 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
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.91)
- Data Roaming and Quality Assessment for Composed Image Retrieval (AAAI 2024 · similarity 0.89)
- Do Composed Image Retrieval Benchmarks Require Multimodal Composition? (NeurIPS 2026 · similarity 0.89)
- ConText-CIR: Learning from Concepts in Text for Composed Image Retrieval (CVPR 2025 · similarity 0.88)
- Reason-before-Retrieve: One-Stage Reflective Chain-of-Thoughts for Training-Free Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.88)
- Context-I2W: Mapping Images to Context-Dependent Words for Accurate Zero-Shot Composed Image Retrieval (AAAI 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.
