CDTFusion

Dataset

The dataset is organized in the following structure:

datasets/
├── fmb/
│   ├── test/
│   │   ├── ir/      # Infrared images
│   │   ├── lbl/     # Segmentation ground truth
│   │   └── vi/      # Visible images
│   └── train/
│       ├── ir/
│       ├── lbl/
│       └── vi/
├── pos/
│   ├── ...
└── whu/
    ├── ...

All images are cropped to a size of 512×512 pixels.

Model Saving Path

Trained models will be saved in the following directory structure:

fusion/
├── fmb/
├── pos/
└── whu/

Training

To train the models for each dataset, use the following commands:

For FMB dataset:

python fmb_train_step1.py
python fmb_train_step2.py

For POS dataset:

python pos_train_step1.py
python pos_train_step2.py

For WHU dataset:

python whu_train_step1.py
python whu_train_step2.py

Testing

To test the trained models, use the following commands:

For FMB dataset:

python fmb_test.py

For POS dataset:

python pos_test.py

For WHU dataset:

python whu_test.py

You can obtain the trained model parameters through this link with password:ndbp.