Research map / Vision-language and multimodal / Human and embodied modeling
Most and Least Retrievable Images in Visual-Language Query Systems
Liuwan Zhu, Rui Ning, Jiang Li, Chunsheng Xin, Hongyi Wu
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 |
|---|---|---|---|
| use APGD-R (random resize, pad and noise) instead of APGD for LRI against JPEG compression (ablation) | LRI · ImageNet, ViT-B/32, JPEG defense | mR@10 secret keyword: 0.4 -> 97.2 | Improved |
| use APGD-R instead of APGD for MRI against JPEG compression (ablation) | MRI · ImageNet, ViT-B/32, JPEG defense | mR@10: 0.4 -> 60.1 | Improved |
| use APGD-R instead of APGD for MRI against random resizing and padding (ablation) | MRI · ImageNet, ViT-B/32, R&P defense | mR@10: 0.2 -> 79.0 | Improved |
| use APGD optimizer instead of PGD for MRI (ablation) | MRI · Paris, ViT-B/32, epsilon 0.03 | mR@1: PGD less than 15% -> APGD over 80% | Improved |
| use APGD instead of PGD for LRI (ablation) | LRI · Paris, ViT-B/32, epsilon 0.02 | mR@1 secret keyword: 80% -> approximately 98% | Improved |
| use FGSM instead of APGD/PGD (ablation) | MRI/LRI · Paris, ViT-B/32 | retrieval probability: MRI top-100 less than 3%, LRI less than 10% | Hurt |
| increase MRI generation iterations from 300 to 1000 on ViT-B/32 (ablation) | MRI · MSCOCO, ViT-B/32 | mR@1: less than 40% -> about 40% top-1; over 87% top-1 on ResNet50 at 1000 | Improved |
| use the proposed LRI loss instead of Fooling targeted attack | LRI · Flickr30k, ResNet50x4 | mR@1 secret keyword: 3.2 -> 97.1 | Improved |
| use the proposed LRI loss instead of Fooling (image quality) | LRI · L2 distortion | L2 distortion: 28.85 -> 8.31 | Improved |
| switch target model from ResNet50 to ViT-B/32 | MRI · ImageNet keyword query | mR@1: 87.3 -> 41.6 | Hurt |
Closest papers by meaning
- Towards Natural Language-Based Document Image Retrieval: New Dataset and Benchmark (CVPR 2025 · similarity 0.83)
- MASS: Overcoming Language Bias in Image-Text Matching (AAAI 2025 · similarity 0.83)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.82)
- MagicLens: Self-Supervised Image Retrieval with Open-Ended Instructions (ICML 2024 · similarity 0.82)
- INQUIRE: A Natural World Text-to-Image Retrieval Benchmark (NeurIPS 2024 · similarity 0.82)
- MRMR: A Realistic and Expert-Level Multidisciplinary Benchmark for Reasoning-Intensive Multimodal Retrieval (ICLR 2026 · similarity 0.82)
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.
