Research map / Vision-language and multimodal / Cross-modal learning
Air-Know: Arbiter-Calibrated Knowledge-Internalizing Robust Network for Composed Image Retrieval
Zhiheng Fu, Yupeng Hu, Qianyun Yang, Shiqi Zhang, Zhiwei Chen, Zixu Li
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 External Prior Arbitration (MLLM anchor labels) (ablation) | Air-Know · FashionIQ val (20% noise) | Avg R@10: 53.99 -> 55.18 | Improved |
| add multi-step cross-verification to MLLM labeling (ablation) | Air-Know · FashionIQ val (20% noise) | Avg R@10: 53.37 -> 55.18 | Improved |
| use geometric deconstruction vector instead of raw triplet features (ablation) | EKI proxy · FashionIQ val (20% noise) | Avg R@10: 53.54 -> 55.18 | Improved |
| add difference and relation vectors to basic (z_q, z_t) input (ablation) | EKI proxy · FashionIQ val (20% noise) | Avg R@10: 54.29 -> 55.18 | Improved |
| add relation vector (z_q * z_t) to GDV (ablation) | EKI proxy · FashionIQ val (20% noise) | Avg R@10: 54.67 -> 55.18 | Improved |
| add difference vector (z_q - z_t) to GDV (ablation) | EKI proxy · FashionIQ val (20% noise) | Avg R@10: 54.47 -> 55.18 | Improved |
| add basic semantics (z_q, z_t) to GDV (ablation) | EKI proxy · FashionIQ val (20% noise) | Avg R@10: 52.80 -> 55.18 | Improved |
| add MC Dropout to EKI (ablation) | EKI proxy · FashionIQ val (20% noise) | Avg R@10: 53.71 -> 55.18 | Improved |
| add dual-stream reconciliation (vs standard InfoNCE) (ablation) | Air-Know · FashionIQ val (20% noise) | Avg R@10: 49.53 -> 55.18 | Improved |
| add clean alignment stream (ablation) | Air-Know · FashionIQ val (20% noise) | Avg R@10: 16.84 -> 55.18 | Improved |
| add feedback reconciliation stream (ablation) | Air-Know · FashionIQ val (20% noise) | Avg R@10: 54.08 -> 55.18 | Improved |
| add feedback reconciliation stream (ablation) | Air-Know · CIRR test (20% noise) | Avg R@K: 79.30 -> 80.24 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- HABIT — HABIT: Chrono-Synergia Robust Progressive Learning Framework for Composed Image Retrieval (AAAI 2026) · better on 3 of 3 shared results
- INTENT — INTENT: Invariance and Discrimination-aware Noise Mitigation for Robust Composed Image Retrieval (AAAI 2026) · better on 3 of 3 shared results
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
- Learning with Noisy Triplet Correspondence for Composed Image Retrieval (CVPR 2025 · similarity 0.87)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.85)
- ConText-CIR: Learning from Concepts in Text for Composed Image Retrieval (CVPR 2025 · similarity 0.85)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.85)
- Do Composed Image Retrieval Benchmarks Require Multimodal Composition? (NeurIPS 2026 · similarity 0.85)
- Image2Sentence based Asymmetrical Zero-shot Composed Image Retrieval (ICLR 2024 · 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.
