Computer Vision News - November 2020

GAN-CIRCLE 13 Diagnostic assessment was also provided by clinicians (certified radiologists with mean experience 12.3 years). An interesting conclusion was that GAN images (as it has been also observed before) provide sharper results with better details of texture, while the PSNR-based methods have higher noise suppression. The visual comparison between the different networks shows the restored bony structures in the red and yellow boxes. To better examine the result of the proposed architecture the zoomed in regions of interest are shown below. Subtle structures do (pointed with colored circles) show better detail than the rest architecture variations. A final interesting finding is that the semi-supervised method had similar performance with the fully supervised methods on the tibia dataset! Very exciting performance and hopefully, more GAN architectures will be explored in the future to gain knowledge out of their compared results. Hope to see you all next month I would really like to thank one of author Chenyu You, for letting me use his ideas and explore them freely in this review, his permission to present and to use the figures and data and his great insight!