Research map / Video / Video segmentation and tracking
RelayFormer: A Unified Local-Global Attention Framework for Scalable Image and Video Manipulation Localization
Wen Huang, Jiarui Yang, Tao Dai, Jiawei Li, Shaoxiong Zhan, Bin Wang, Shu-Tao Xia
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 GLRA with 1 GLR token (ablation) | Relay-ViT · image manipulation avg F1 | F1: 0.454 -> 0.521 | Improved |
| add query-based mask decoder (n=1) (ablation) | Relay-ViT · image manipulation avg F1 | F1: 0.521 -> 0.532 | Improved |
| use 2 GLR tokens instead of 1 (ablation) | Relay-ViT · image manipulation avg F1 | F1: 0.532 -> 0.554 | Improved |
| use 3 GLR tokens instead of 2 (ablation) | Relay-ViT · image manipulation avg F1 | F1: 0.554 -> 0.524 | Hurt |
| add 4D RoPE to GLRA (ablation) | Relay-ViT · image manipulation avg F1 | F1: 0.517 -> 0.554 | Improved |
| add temporal GLRA on top of spatial GLRA (ablation) | Relay · MOSE100 | F1: 0.6745 -> 0.6877 | Improved |
| add spatial GLRA (vs none) (ablation) | Relay · MOSE100 | F1: 0.6124 -> 0.6745 | Improved |
| avoid resizing 4K inputs to 1024x1024 (ablation) | Relay · IMD2020 | F1: 0.350 -> 0.453 | Improved |
| use 512 sub-images instead of 256 (with overlap) (ablation) | Relay-ViT · image manipulation avg F1 | F1: 0.477 -> 0.554 | Improved |
| add overlap between 256 sub-images (ablation) | Relay-ViT · image manipulation avg F1 | F1: 0.469 -> 0.477 | Improved |
| apply GLRA in all layers instead of sparse layers (ablation) | Relay-ViT · CASIAv1 | F1 (%): 73.24 -> 75.50 | Improved |
| add video forgery data to image training (ablation) | Relay-Seg · image benchmarks | F1: COV 0.551 -> 0.569; NIST16 0.335 -> 0.282 | No clear effect |
Closest papers by meaning
- UnionFormer: Unified-Learning Transformer with Multi-View Representation for Image Manipulation Detection and Localization (CVPR 2024 · similarity 0.90)
- Collaborative Transformers with Multi-Level Forensic Attention for Image Manipulation Localization (AAAI 2026 · similarity 0.89)
- Towards Modern Image Manipulation Localization: A Large-Scale Dataset and Novel Methods (CVPR 2024 · similarity 0.87)
- Training-Free Image Manipulation Localization Using Diffusion Models (AAAI 2025 · similarity 0.86)
- GIM: A Million-scale Benchmark for Generative Image Manipulation Detection and Localization (AAAI 2025 · similarity 0.86)
- ForensicHub: A Unified Benchmark & Codebase for All-Domain Fake Image Detection and Localization (NeurIPS 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.
