SARDINet: Impact of the Decoding Strategy

This repository is dedicated to the impact of the decoding strategy on the optical reconstruction of SARDINet (from [Bralet et al, 2022]) from a SAR image. We implemented here five different decoding strategies :

1. Post-upsampling convolutions
2. Transposed convolutions
3. Sub-pixel convolutions from [Shi et al, 2016]
4. Post-upsampling convolution with a last sub-pixel convolution layer
5. Transposed convolutions with a last sub-pixel convolution layer

More details can be found in the article Impact de la stratégie de décodage sur la traduction de modalité radar-optique d'images de télédétection

The code was computed using Python 3.10.6 and the packages versions detailed in the file requirements.txt.

In order to run the code please follow the next steps :

• Open the main.py file
• Choose the path where your data are located and the path where you want to save the results
• Choose your hyperparameters, the number of the decoder (as mentionned above) and your loss functions
• If you want to modify the architecture of the network, you'll find all you need in the TransNet.py file.
• Save your changes
• Run the main.py file

Citation

If this work was useful for you, please ensure citing our works :

Bralet, A., Atto, A., Chanussot, J., and Trouve, E. (2023). Impact de la stratégie de décodage sur la traduction de modalité radar-optique d’images de télédétection. In 29° Colloque sur le traitement du signal et des images, number 2023-1309, pages p. 929–932, Grenoble. GRETSI - Groupe de Recherche en Traitement du Signal et des Images

Thank you for your support

Any troubles ?

If you have any troubles with the article or the code, do not hesitate to contact us !

References

[Bralet et al, 2022] A. Bralet, A. M. Atto, J. Chanussot and E. TrouvÉ, "Deep Learning of Radiometrical and Geometrical Sar Distorsions for Image Modality translations," 2022 IEEE International Conference on Image Processing (ICIP), Bordeaux, France, 2022, pp. 1766-1770, doi: 10.1109/ICIP46576.2022.9897713.

[Shi et al, 2016] W. Shi et al., "Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016, pp. 1874-1883, doi: 10.1109/CVPR.2016.207.