Research map / Vision-language and multimodal / Human and embodied modeling
MAGIC: Multimodal relAtional Graph adversarIal inferenCe for Diverse and Unpaired Text-Based Image Captioning
Wenqiao Zhang, Haochen Shi, Jiannan Guo, Shengyu Zhang, Qingpeng Cai, Juncheng Li, Sihui Luo, Yueting Zhuang
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 MAGIC instead of ISM-LSTM | Text-based image captioning · TextCaps validation | CIDEr: 31.1 -> 76.6 | Improved |
| add a central object-aware pool (ablation) | MAGIC · TextCaps validation | CIDEr: 71.3 -> 76.6 | Improved |
| add query attention over text tokens (ablation) | MAGIC · TextCaps validation | CIDEr: 75.9 -> 76.6 | Improved |
| use CGAN rather than the original GAN for alignment (ablation) | MAGIC · TextCaps validation | CIDEr: 65.3 -> 76.6 | Improved |
| add the language discriminator (ablation) | MAGIC · TextCaps validation | CIDEr: 73.9 -> 76.6 | Improved |
| vary COP capacity from N_k=3 to N_k=1 (ablation) | MAGIC · TextCaps validation | CIDEr: 76.6 -> 77.3 | Mixed |
| vary COP capacity from N_k=3 to N_k=2 (ablation) | MAGIC · TextCaps validation | CIDEr: 76.6 -> 75.8 | No clear effect |
| vary COP capacity from N_k=3 to N_k=4 (ablation) | MAGIC · TextCaps validation | CIDEr: 76.6 -> 75.3 | No clear effect |
| vary COP capacity from N_k=3 to N_k=5 (ablation) | MAGIC · TextCaps validation | CIDEr: 76.6 -> 73.4 | No clear effect |
| select central objects by learning their scores rather than using center, large-object, or random rules (ablation) | MAGIC · TextCaps validation | CIDEr: 72.3, 75.9, 67.8 -> 76.6 | Improved |
Later papers that compared with this method
8 later accepted papers report MAGIC as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- CHARM: Collaborative Harmonization Across Arbitrary Modalities for Modality-Agnostic Semantic Segmentation (AAAI 2026 · better on 10 of 10 shared results)
- Towards Robust Multi-Modal Semantic Segmentation with Teacher-Student Framework and Hybrid Prototype Distillation (CVPR 2026 · better on 4 of 4 shared results)
- VLBiMan: Vision-Language Anchored One-Shot Demonstration Enables Generalizable Bimanual Robotic Manipulation (ICLR 2026 · better on 3 of 3 shared results)
- VPN: Visual Prompt Navigation (AAAI 2026 · better on 1 of 1 shared result)
- Reducing Unimodal Bias in Multi-Modal Semantic Segmentation with Multi-Scale Functional Entropy Regularization (ICCV 2025 · better on 6 of 7 shared results)
- Multi-Task Domain Adaptation for Language Grounding with 3D Objects (ECCV 2024 · better on 2 of 4 shared results)
- ZeroNLG: Aligning and Autoencoding Domains for Zero-Shot Multimodal and Multilingual Natural Language Generation (TPAMI 2024 · better on 1 of 1 shared result)
- Socratic Models: Composing Zero-Shot Multimodal Reasoning with Language (ICLR 2023 · better on 0 of 1 shared result)
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
- Adversarial Semantic Alignment for Improved Image Captions (CVPR 2019 · similarity 0.87)
- Dense Relational Captioning: Triple-Stream Networks for Relationship-Based Captioning (CVPR 2019 · similarity 0.87)
- Towards Accurate Text-Based Image Captioning With Content Diversity Exploration (CVPR 2021 · similarity 0.87)
- CapsFusion: Rethinking Image-Text Data at Scale (CVPR 2024 · similarity 0.86)
- Relational Distant Supervision for Image Captioning without Image-Text Pairs (AAAI 2024 · similarity 0.86)
- GRAPHGPT-O: Synergistic Multimodal Comprehension and Generation on Graphs (CVPR 2025 · similarity 0.85)
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.
