Research map / Vision-language and multimodal / Human and embodied modeling
Adapting In-context Generation for Enhanced Composed Image Retrieval
Haiwen Li, Zining Chen, Delong Liu, Zhaohui Hou, Zhicheng Zhao, Fei Su
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 DAIG synthetic pretraining + FRA to SPRC at 32-shot | SPRC · CIRR test | Recall@5: 57.61 -> 72.41 | Improved |
| add DAIG to BLIP4CIR at 32-shot | BLIP4CIR · CIRR test | Recall@5: 27.86 -> 57.01 | Improved |
| add DAIG to SPRC with full labels | SPRC · CIRR test | Recall@5: 82.22 -> 84.10 | Improved |
| add DAIG to SPRC with full labels | SPRC · FashionIQ val | Average Recall@50: 74.95 -> 75.96 | Improved |
| add DAIG to CLIP4CIR at 1% labels on FashionIQ Dress | CLIP4CIR · FashionIQ val Dress | Recall@10: 24.79 -> 24.74 | No clear effect |
| use DAIG triplets vs PTG for few-shot SPRC (8-shot) | SPRC · FashionIQ | Avg@10: 30.6 -> 43.7 | Improved |
| train SPRC on 20k DAIG triplets instead of 534k CIRHS (ablation) | SPRC · CIRR | Avg.: 69.14 -> 71.68 | Improved |
| use domain-adapted LoRA T2I instead of zero-shot in-context generation (ablation) | SPRC · CIRR | Recall@5: 66.10 -> 70.19 | Improved |
| replace LoRA with CIR-LoRA (MoE) (ablation) | SPRC · CIRR | Recall@5: 70.19 -> 71.01 | Improved |
| add visual feature perturbation (p=0.5) in DRSP (ablation) | SPRC · CIRR | Recall@5: 71.01 -> 72.02 | Improved |
| add angular margin in FRA (ablation) | SPRC · CIRR | Recall@5: 74.65 -> 75.52 | Improved |
| increase number of synthetic triplets beyond 30k (ablation) | SPRC + DAIG · CIRR/FashionIQ | Recall@5 | No clear effect |
Closest papers by meaning
- ConText-CIR: Learning from Concepts in Text for Composed Image Retrieval (CVPR 2025 · similarity 0.94)
- Generative Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.92)
- Visual Delta Generator with Large Multi-modal Models for Semi-supervised Composed Image Retrieval (CVPR 2024 · similarity 0.92)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.91)
- Vision-by-Language for Training-Free Compositional Image Retrieval (ICLR 2024 · similarity 0.91)
- Context-I2W: Mapping Images to Context-Dependent Words for Accurate Zero-Shot Composed Image Retrieval (AAAI 2024 · similarity 0.91)
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.
