Research map / Image generation / Image restoration
MMAIF: Multi-task and Multi-degradation All-in-One for Image Fusion with Language Guidance
Zihan Cao, Yu Zhong, Ziqi Wang, Liang-Jian Deng
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 MoE GLU in place of dense FFN (ablation) | MMAIF DiT (FM) · LLVIP degraded VIF | VIF: 0.436 -> 0.582 | Improved |
| use DiT backbone instead of U-net at similar parameter count (ablation) | MMAIF (FM) · LLVIP degraded VIF | VIF: 0.423 -> 0.436 | Improved |
| add 2D RoPE and per-block absolute PE (ablation) | MMAIF DiT · LLVIP degraded VIF | VIF: 0.582 -> 0.593 | Improved |
| add attention value residual (ablation) | MMAIF DiT · LLVIP degraded VIF | VIF: 0.593 -> 0.599 | Improved |
| add NAFBlock convolution before attention (ablation) | MMAIF DiT · LLVIP degraded VIF | VIF: 0.599 -> 0.607 | Improved |
| add LoRA-AdaLN conditioning (ablation) | MMAIF DiT · LLVIP degraded VIF | VIF: 0.607 -> 0.604 | No clear effect |
| use 100 Euler steps instead of 25 (ablation) | MMAIF (FM) · LLVIP degraded VIF | VIF: 0.592 -> 0.605 | Improved |
| use Gaussian prior p0 instead of degraded latent Z0 (ablation) | MMAIF (FM) · LLVIP degraded VIF | VIF: 0.435 -> 0.604 | Improved |
| use stochastic I2SB sampling with p0=mean instead of deterministic p0=mean (ablation) | MMAIF (FM) · LLVIP degraded VIF | VIF: 0.438 -> 0.601 | Improved |
| use Cosmos VAE tokenizer instead of Flux KL-VAE (ablation) | MMAIF latent space · SICE+LLVIP reconstruction | PSNR: 33.41 -> 34.02 | Improved |
| use flow matching instead of regression | MMAIF · LLVIP degraded VIF | VIF: 0.556 -> 0.604 | Improved |
| use all-in-one MMAIF instead of DA-CLIP+EMMA restoration+fusion under three simultaneous degradations | MMAIF · LLVIP multi-degradation (n=3) | VIF: 0.431 -> 0.584 | Improved |
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.
- TextIF — Text-IF: Leveraging Semantic Text Guidance for Degradation-Aware and Interactive Image Fusion (CVPR 2024) · better on 2 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
- ControlFusion: A Controllable Image Fusion Network with Language-Vision Degradation Prompts (NeurIPS 2025 · similarity 0.88)
- Image Fusion via Vision-Language Model (ICML 2024 · similarity 0.88)
- MMAR: Towards Lossless Multi-Modal Auto-Regressive Probabilistic Modeling (CVPR 2025 · similarity 0.87)
- One Model for ALL: Low-Level Task Interaction Is a Key to Task-Agnostic Image Fusion (CVPR 2025 · similarity 0.87)
- Task-driven Image Fusion with Learnable Fusion Loss (CVPR 2025 · similarity 0.87)
- Efficient Rectified Flow for Image Fusion (NeurIPS 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.
