Research map / Vision-language and multimodal / Cross-modal learning
Bridging Pixels and Words: Mask-Aware Local Semantic Fusion for Multimodal Media Verification
Zizhao Chen, Ping Wei, Ziyang Ren, Huan Li, Xiangru Yin
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 mask-label pairs (paired local semantics) as anchors (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Binary ACC (%): 88.93 -> 89.37 | Improved |
| add mask-label pairs (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Image grounding IoUmean (%): 76.12 -> 82.37 | Improved |
| use text labels only as local semantics (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Text grounding F1 (%): 75.59 -> 76.69 | Improved |
| use image masks only as local semantics (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Image grounding IoUmean (%): 76.12 -> 82.02 | Improved |
| use bidirectional verification instead of text-only or image-only query verification (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Binary ACC (%): 88.95 -> 89.37 | Improved |
| add BCV verification on top of plain image and text branches (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Binary ACC (%): 88.03 -> 89.37 | Improved |
| add multi-label shallow and deep fusion (MSF, MDF) instead of fusion by addition (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Binary ACC (%): 87.97 -> 89.37 | Improved |
| add gate weight module (ablation) | MaLSF · DGM4 (Open Vocabulary Parser) | Image manipulation grounding | Improved |
| use Open Vocabulary Parser instead of Caption-Anchored Parser | MaLSF · DGM4 | Text grounding F1 (%): 76.88 -> 77.19 | Improved |
| choice of parser (Open Vocabulary vs Caption-Anchored) | MaLSF · Weibo17 | accuracy (%): 93.4 vs 93.5 | No clear effect |
| use MaLSF instead of EMSF | MaLSF · DGM4 | Image grounding IoU50 / IoUmean: 80.83 -> 82.37 (IoUmean) | Improved |
| use MaLSF instead of EMSF | MaLSF · DGM4 | Multi-label mAP (%): 91.42 -> 90.76 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HAMMER++ — HAMMER: Harnessing MLLMs via Cross-Modal Integration for Intention-Driven 3D Affordance Grounding (CVPR 2026) · better on 4 of 4 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
- Unleashing the Potential of Consistency Learning for Detecting and Grounding Multi-Modal Media Manipulation (CVPR 2025 · similarity 0.89)
- DAMMFND: Domain-Aware Multimodal Multi-view Fake News Detection (AAAI 2025 · similarity 0.87)
- Reinforced Adaptive Knowledge Learning for Multimodal Fake News Detection (AAAI 2024 · similarity 0.86)
- Text-Guided Fine-grained Counterfactual Inference for Short Video Fake News Detection (AAAI 2025 · similarity 0.86)
- MMFakeBench: A Mixed-Source Multimodal Misinformation Detection Benchmark for LVLMs (ICLR 2025 · similarity 0.86)
- CrossCheck-Bench: Diagnosing Compositional Failures in Multimodal Conflict Resolution (AAAI 2026 · similarity 0.86)
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.
