TTFF322 / GCEANet

The source code of the paper: Side-Scan Sonar Image Recognition with Zero-Shot and Style Transfer

GCEANet

The source code of the paper: Side-Scan Sonar Image Recognition with Zero-Shot and Style Transfer

Requirements

We recommend the following configurations:

• python 3.8
• PyTorch 1.8.0
• CUDA 11.1

The model code will be published soon.

Model Training

• Download the content dataset: MS-COCO.
• Download the style dataset: WikiArt.
• Download the pre-trained VGG-19 model.
• Put your trained model to ./model/ folder.
• Set your available GPU ID in Line 92 of the file "train.py".
• Run the following command:

python train.py --content_dir /data/train2014 --style_dir /data/WikiArt/train

pseudo-SSS image synthesis

• All optical images of the sonar style transfer test can be downloaded: optical-content.
• Put the optical images to ./Optic_test/ folder.
• The sonar images as style images used for pseudo-SSS image synthesis are at *./sonar_test.
• Run the following command:

python batch_convert.py

The weights of our model can be found:weights.

Real SSS image for classification test.

• The real sonar images can be downloaded: real_dataset and KLSG.

Acknowledgments

The code in this repository is based on Li et al. and SANet. Thanks for both their paper and code.