[CVPR 2025] ARGS-Diff

Self-Learning Hyperspectral and Multispectral Image Fusion via Adaptive Residual Guided Subspace Diffusion Model

Jian Zhu, He Wang, Yang Xu, Zebin Wu, and Zhihui Wei

Nanjing University of Science and Technology

Framework

Requirements

1. Environment setup

conda create -n args python=3.9
conda activate args

2. Requirements installation

pip install -r requirements.txt

Quick Start

python sample_subspace.py --mode 'semi'

Sample on Your Own Data

Train

refer to ARGS-Diff-train to train the spatial and spectral networks

Sample

1. Place the pavia.mat file into the data folder. This file should contain the following keys: LR-HSI, HR-MSI, and optionally HR-HSI.

2. Copy the pretrained model file ema_0.9999_030000.pt from the training project ARGS-Diff-train spatial_train_result/pavia/ to the ckpt/pavia/ directory of the current project, and rename it to spa.pt.

3. Modify line 46 in sample_subspace.py to use "pavia", then run:

   python sample_subspace.py --mode 'semi'

Acknowledge

Some of the codes are built upon denoising-diffusion-pytorch and MIAE.