@article{4064,
  author = {Francesca Fiani, Adriano Puglisi and Christian Napoli},
  title = {The Cross-depiction of Images through Neural Style Transfer},
  journal = {Journal of Data Processing},
  year = {2024},
  volume = {14},
  number = {2},
  doi = {https://doi.org/10.6025/jdp/2024/14/2/47-55},
  url = {https://www.dline.info/jdp/fulltext/v14n2/jdpv14n2_1.pdf},
  abstract = {Current models for computer vision in neural networks are trained using billions of images. The concept is that these models can improve their generalisation ability when the dataset includes a wide range of images, such as those with different lighting and environmental conditions of the same objects. This ability to generalize is essential in the task of object detection, particularly when it comes to the challenge of cross-depiction. In this research, we investigate the application of Neural Style Transfer as a new method to transform the initial data and improve model generalization. To test the impact on performance for object detection models, we chose the Faster R-CNN model for application on the Pascal VOC 2012 dataset. Several experiments were conducted, including style changes on images and adjustments to the parameters of Neural Style Transfer to preserve the content of the original images. The results were encouraging, suggesting a promising basis for further research on cross-depiction through Neural Style Transfer. 
},
}