Research map / Vision-language and multimodal / CLIP and zero-shot
G-MIXER: Geodesic Mixup-based Implicit Semantic Expansion and Explicit Semantic Re-ranking for Zero-Shot Composed Image Retrieval
Jiyoung Lim, Heejae Yang, Jee-Hyong Lee
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 Delta (include/exclude caption similarity difference) to re-ranking score (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 22.49 -> 28.29 | Improved |
| add S_lambda to re-ranking score (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 16.43 -> 28.29 | Improved |
| add S_lambda and Delta together (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 11.80 -> 28.29 | Improved |
| add S_m (modification text similarity) to re-ranking score (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 24.30 -> 28.29 | Improved |
| add geodesic mixup (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 24.77 -> 28.29 | Improved |
| add geodesic mixup (ablation) | G-MIXER (ViT-L/14) · CIRR test | Recall@1: 33.69 -> 37.42 | Improved |
| add geodesic mixup (ablation) | G-MIXER (ViT-L/14) · FashionIQ val (average) | Recall@10: 34.36 -> 41.92 | Improved |
| use full Delta rather than Delta_in only (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 21.17 -> 28.29 | Improved |
| use full Delta rather than Delta_ex only (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 17.57 -> 28.29 | Improved |
| use a range of mixup ratios instead of a fixed ratio (ablation) | G-MIXER · CIRCO, CIRR, FashionIQ | retrieval accuracy | Improved |
| use GPT-4o instead of GPT-4o-mini (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 27.18 -> 28.29 | No clear effect |
| use GPT-4o instead of GPT-turbo (ablation) | G-MIXER (ViT-L/14) · CIRCO test | mAP@5: 26.68 -> 28.29 | No clear effect |
Closest papers by meaning
- Language-only Training of Zero-shot Composed Image Retrieval (CVPR 2024 · similarity 0.87)
- Generative Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.87)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.87)
- Vision-by-Language for Training-Free Compositional Image Retrieval (ICLR 2024 · similarity 0.86)
- Reason-before-Retrieve: One-Stage Reflective Chain-of-Thoughts for Training-Free Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.86)
- Imagine and Seek: Improving Composed Image Retrieval with an Imagined Proxy (CVPR 2025 · similarity 0.86)
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.
