[New Publication] TCL in CVIU

It is a great pleasure to announce that our Computer Vision and Image Understanding (CVIU) paper describing our novel Translation-Classification Loss is now available online!

This Translation-Classification Loss (TCL) aims at improving the quality of SAR-to-optical translations by leveraging a frozen auxiliary ResNet50 classifier network to evaluate the quality of the translated optical image. The TCL is tested on 10 different architectures among which we introduce an improvement of our previous SARDINet architecture [1], named SARDINetV2. The use of the TCL significantly enhances not only the credibility of the images but it especially increases the reliability of the reconstructed patterns and avoid unreliable artifacts. All experiments were conducted on BigEarthNet-MM dataset [2].

You can access the CVIU paper directly on ScienceDirect and on HAL.

If our work can be of any use for use, please consider citing it as:

Translation-classification loss for SAR image understanding with deep learning, Bralet, A., Atto, A. M., Chanussot, J., & Trouvé, E., in Computer Vision and Image Understanding, 2025, Elsevier, p. 104374, doi: 10.1016/j.cviu.2025.104374.

Reference:

[1] Bralet, A., Atto, A. M., Chanussot, J., & Trouvé, E. (2022, October), “Deep Learning of Radiometrical and Geometrical Sar Distorsions for Image Modality translations”, in 2022 IEEE International Conference on Image Processing (ICIP) (pp. 1766-1770). Bordeaux, France. IEEE

[2] G. Sumbul et al., “BigEarthNet-MM: A Large-Scale, Multimodal, Multilabel Benchmark Archive for Remote Sensing Image Classification and Retrieval [Software and Data Sets],” in IEEE Geoscience and Remote Sensing Magazine, vol. 9, no. 3, pp. 174-180, Sept. 2021, doi: 10.1109/MGRS.2021.3089174.