Research map / Vision-language and multimodal / Human and embodied modeling
PinPoint: Evaluation of Composed Image Retrieval with Explicit Negatives, Multi-Image Queries, and Paraphrase Testing
Rohan Mahadev, Joyce Yuan, Patrick Poirson, David Xue, Hao-Yu Wu, Dmitry Kislyuk
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 training-free Qwen2.5-VL-7B reranking | Meta CLIP 2 Combined · PinPoint | mAP@10: 0.044 -> 0.087 | Improved |
| add training-free Qwen2.5-VL-7B reranking | MMRet-CLIP-L · PinPoint | mAP@10: 0.178 -> 0.236 | Improved |
| add training-free Qwen2.5-VL-7B reranking | MagicLens-CLIP-L · PinPoint | mAP@10: 0.155 -> 0.231 | Improved |
| add training-free Qwen2.5-VL-7B reranking | GPT-5-Text Premerge · PinPoint | mAP@10: 0.266 -> 0.272 | Improved |
| add training-free Qwen2.5-VL-7B reranking | MMRet-MLLM-S1 · PinPoint | mAP@10: 0.224 -> 0.290 | Improved |
| add MLLM reranking | MMRet-MLLM-S1 · PinPoint | NegRecall@10: 0.091 -> 0.056 | Improved |
| add MLLM reranking | MMRet-MLLM-S1 · PinPoint | Delta mAP (%): 6.38 -> 2.01 | Improved |
| add MLLM reranking | MagicLens-CLIP-L · PinPoint | linguistic sensitivity range: 0.182 -> 0.201 | Hurt |
| add MLLM reranking | Meta CLIP 2 Combined · PinPoint | linguistic sensitivity range: 0.115 -> 0.150 | Hurt |
| add MLLM reranking | MMRet-MLLM-S1 · PinPoint multi-image queries | mAP@10: 0.067 -> 0.067 | No clear effect |
| use multi-image rather than single-image queries | MMRet-MLLM-S1 · PinPoint | mAP@10: 0.324 -> 0.067 | Hurt |
| add composed image-plus-text fusion versus image-only embedding | Meta CLIP 2 · PinPoint | mAP@10: 0.017 -> 0.044 | Improved |
Closest papers by meaning
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.93)
- Reason-before-Retrieve: One-Stage Reflective Chain-of-Thoughts for Training-Free Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.90)
- 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 0.89)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.89)
- Generative Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.89)
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.
