Research map / Vision-language and multimodal / CLIP and zero-shot
Visual Clues: Bridging Vision and Language Foundations for Image Paragraph Captioning
Yujia Xie, Luowei Zhou, Xiyang Dai, Lu Yuan, Nguyen Bach, Ce Liu, Michael Zeng
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 structured visual clues with VG-domain tags instead of BLIP-large | Image paragraph captioning · Stanford dataset test | SPIPE F-score: 7.6 -> 10.0 | Improved |
| Use structured visual clues with VG-domain tags instead of Socratic model | Image paragraph captioning · Stanford dataset test | SPIPE F-score: 3.2 -> 10.0 | Improved |
| Specify the tag vocabulary to the VG domain rather than using general-domain tags | BEST · Stanford dataset test | SPIPE F-score: 8.8 -> 10.0 | Improved |
| Use visual clues rather than Socratic prompts for VQA | Visual question answering · GQA (Generative) | accuracy: 24.95 -> 37.00 | Improved |
| Use visual clues rather than Socratic prompts for VQA | Visual question answering · GQA (Discriminative) | accuracy: 26.89 -> 39.93 | Improved |
| Use visual clues rather than Socratic prompts for VQA | Visual question answering · OK-VQA (Generative) | accuracy: 16.50 -> 28.89 | Improved |
| Add local object descriptions (ablation) | BEST-VG domain · Stanford dataset test | SPIPE F-score: 8.0 -> 10.0 | Improved |
| Add the caption model (ablation) | BEST-VG domain · Stanford dataset test | SPIPE F-score: 8.7 -> 10.0 | Improved |
| Add local descriptions and captions alongside tags (ablation) | BEST-VG domain · Stanford dataset test | SPIPE F-score: 5.9 -> 10.0 | Improved |
| Use the open-vocabulary extractor rather than YOLOv5 (ablation) | BEST-VG domain · Stanford dataset test | SPIPE F-score: 9.0 -> 10.0 | Improved |
| Use Florence-H rather than CLIP ViT-L/14 as the tagger (ablation) | BEST-VG domain · Stanford dataset test | SPIPE F-score: 7.8 -> 10.0 | Improved |
| Use Davinci-text-001 rather than the smaller Curie language model (ablation) | BEST-VG domain · Stanford dataset test | SPIPE F-score: 8.9 -> 10.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- BLIP-large — BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (ICML 2022) · better on 1 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.88)
- Benchmarking Large Vision-Language Models via Directed Scene Graph for Comprehensive Image Captioning (CVPR 2025 · similarity 0.88)
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.88)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (NeurIPS 2025 · similarity 0.87)
- VisMin: Visual Minimal-Change Understanding (NeurIPS 2024 · similarity 0.87)
- What's in the Image? A Deep-Dive into the Vision of Vision Language Models (CVPR 2025 · similarity 0.87)
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.
