CtrlFuse: Mask-Prompt Guided Controllable Infrared and Visible Image Fusion (Official PyTorch Implementation)

January 30, 2026 · View on GitHub

This repository contains the official PyTorch implementation of the paper: "CtrlFuse: Mask-Prompt Guided Controllable Infrared and Visible Image Fusion" (Accepted by AAAI 2026)

Authors: Yiming Sun, Yuan Ruan, Qinghua Hu,Pengfei Zhu Affiliation: VisDrone Group

📢 News

[2026-01]: Code and pre-trained models are released!
[2025-11-08]: The paper is accepted by AAAI 2026.

Infrared and visible image fusion generates all-weather perception-capable images by combining complementary modalities, enhancing environmental awareness for intelligent unmanned systems. Existing methods either focus on pixel-level fusion while overlooking downstream task adaptability or implicitly learn rigid semantics through cascaded detection/segmentation models, unable to interactively address diverse semantic target perception needs. We propose CtrlFuse, a controllable image fusion framework that enables interactive dynamic fusion guided by mask prompts. The model integrates a multi-modal feature extractor, a reference prompt encoder (RPE), and a prompt-semantic fusion module(PSFM). The RPE dynamically encodes task-specific semantic prompts by fine-tuning pre-trained segmentation models with input mask guidance, while the PSFM explicitly injects these semantics into fusion features. Through synergistic optimization of parallel segmentation and fusion branches, our method achieves mutual enhancement between task performance and fusion quality. Experiments demonstrate state-ofthe-art results in both fusion controllability and segmentation accuracy, with the adapted task branch even outperforming the original segmentation model.

Network Architecture Figure 1: The overall architecture of our proposed CtrlFuse.

🔨 Requirements

The code has been tested with Python 3.8 and PyTorch 2.0.0 .

Checkpoints can be downloaded with the links below:

[Baidu Yun]

Additionally, you can download the ViT-H SAM model from the official Segment-anything website:

[Segment-anything]

# 1. Create a conda environment
conda create -n ctrlfuse python=3.8
conda activate ctrlfuse

# 2. Install dependencies
pip install -r requirements.txt

# 3.  Segment-Anything-Model setting
cd ./segment-anything
pip install -v -e .
cd ..

📂 Data Preparation

Please organize your dataset as follows. Note: Ensure that the Visible and Infrared images are strictly aligned (registered) and have the same filenames.

Project_Root/
├── dataset/
│   ├── train/
│   │   ├── vi/             # Visible images (RGB)
│   │   │   ├── 1.jpg
│   │   │   └── ...
│   │   └── ir/           # Infrared images (Grayscale)
│   │   │   ├── 1.jpg
│   │   │   └── ...
│   │   └── mask/             # mask (Grayscale)
│   │       ├── 1.jpg
│   │       └── ...
│   └── test/
│   │   ├── vi/             # Visible images (RGB)
│   │   │   ├── 1.jpg
│   │   │   └── ...
│   │   └── ir/           # Infrared images (Grayscale)
│   │   │   ├── 1.jpg
│   │   │   └── ...
│   │   └── mask/             # mask (Grayscale)
│   │       ├── 1.jpg
│   │       └── ...

🚀 Usage

📊 Results

FMB Dataset	MSE	PSNR	Q_abf	N_abf	SSIM	SCD
LDFusion	0.061	60.71	0.51	0.112	0.514	1.549
SwinFuse	0.042	62.334	0.577	0.029	0.905	1.900
NestFuse	0.046	61.96	0.483	0.042	0.787	1.594
CDDFuse	0.048	62.696	0.674	0.026	1.002	1.626
DIDFuse	0.047	61.565	0.528	0.042	0.765	1.824
SeAFusion	0.047	62.539	0.654	0.029	0.964	1.62
PSFusion	0.051	61.517	0.627	0.056	0.836	1.875
SDCFusion	0.048	62.456	0.693	0.031	0.906	1.657
CtrlFuse(Ours)	0.043	63.292	0.719	0.024	0.925	1.522

Drone Vehicle Dataset	MSE	PSNR	Q_abf	N_abf	SSIM	SCD
LDFusion	0.076	59.573	0.376	0.054	0.568	1.38
SwinFuse	0.084	59.165	0.202	0.069	0.558	1.295
NestFuse	0.071	59.786	0.307	0.052	0.486	1.413
CDDFuse	0.065	60.199	0.469	0.021	0.845	1.359
DIDFuse	0.067	59.988	0.265	0.062	0.466	1.459
SeAFusion	0.094	58.649	0.492	0.044	0.879	1.472
PSFusion	0.067	60.065	0.454	0.095	0.717	1.534
SDCFusion	0.078	59.443	0.534	0.035	0.853	1.316
CtrlFuse(Ours)	0.063	60.317	0.496	0.035	0.779	1.552

MSRS Dataset	MSE	PSNR	Q_abf	N_abf	SSIM	SCD
LDFusion	0.056	61.05	0.438	0.116	0.541	1.515
SwinFuse	0.038	63.69	0.178	0.026	0.343	1.033
NestFuse	0.033	64.128	0.242	0.025	0.217	1.138
CDDFuse	0.038	64.309	0.689	0.023	1.001	1.623
DIDFuse	0.035	63.94	0.204	0.025	0.223	1.121
SeAFusion	0.036	64.491	0.675	0.021	0.982	1.707
PSFusion	0.037	64.001	0.676	0.042	0.917	1.812
SDCFusion	0.039	64.003	0.712	0.023	0.957	1.739
CtrlFuse(Ours)	0.035	64.75	0.685	0.018	0.969	1.726

🤝 Citation

📧 Contact

If you have any other questions about the code, please email ruanyuan@seu.edu.cn.