Research map / Vision-language and multimodal / Human and embodied modeling
Image Retrieval on Real-Life Images With Pre-Trained Vision-and-Language Models
Zheyuan Liu, Cristian Rodriguez-Opazo, Damien Teney, Stephen Gould
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 OSCAR vision-and-language pre-trained initialization (ablation) | CIRPLANT · CIRR test | Recall@1: 15.18 -> 19.55 | Improved |
| add OSCAR vision-and-language pre-trained initialization (ablation) | CIRPLANT · CIRR test | Recall@5: 43.36 -> 52.55 | Improved |
| add OSCAR vision-and-language pre-trained initialization (ablation) | CIRPLANT · CIRR test | Recall_Subset@1: 33.81 -> 39.20 | Improved |
| add OSCAR vision-and-language pre-trained initialization (ablation) | CIRPLANT · Fashion-IQ | (R@10+R@50)/2: 25.17 -> 30.20 | Improved |
| use transformer composition (CIRPLANT without init) instead of TIRG gating | TIRG · CIRR test | Recall_Subset@1: 22.67 -> 33.81 | Improved |
| use transformer composition (CIRPLANT without init) instead of TIRG gating | TIRG · CIRR test | Recall@1: 14.61 -> 15.18 | No clear effect |
| add BERT pre-trained word representations (ablation) | MAAF · CIRR test | (R@5+R_Subset@1)/2: 27.04 -> 27.57 | No clear effect |
| remove output of text tokens in final pooling (MAAF-IT) (ablation) | MAAF · CIRR test | (R@5+R_Subset@1)/2: 27.04 -> 27.02 | No clear effect |
| replace final resolution-wise pooling with average pooling (MAAF-RP) (ablation) | MAAF · CIRR test | (R@5+R_Subset@1)/2: 27.04 -> 27.37 | No clear effect |
| inject text at last convolution layer instead of last FC layer (LastConv) (ablation) | TIRG · CIRR test | (R@5+R_Subset@1)/2: 35.52 -> 29.75 | Hurt |
| compose image and text with TIRG instead of using image only | Image-only baseline · CIRR test | Recall@5: 48.46 -> 48.37 | No clear effect |
| use text only instead of image-text composition (Recall_Subset) | Image+Text Concatenation · CIRR test | Recall_Subset@1: 23.74 -> 39.69 | Mixed |
Closest papers by meaning
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.92)
- Vision-by-Language for Training-Free Compositional Image Retrieval (ICLR 2024 · similarity 0.91)
- ConText-CIR: Learning from Concepts in Text for Composed Image Retrieval (CVPR 2025 · similarity 0.91)
- Generative Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.90)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.90)
- Reason-before-Retrieve: One-Stage Reflective Chain-of-Thoughts for Training-Free Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.90)
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.
