Skip to main content
. 2021 Mar 19;25(6):1864–1872. doi: 10.1109/JBHI.2021.3067465

Fig. 1.

Fig. 1.

An overview of the proposed weakly supervised COVID-19 lesion localization. (a) During model training, a generator and a discriminator work together to remove potential lesions. The image quality of the generated fake normal images is boosted by feature match. (b) During model inference, we obtained the localization map by subtracting output from its input of the generator. The localization map is added to the original image to augment the COVID-19 diagnosis.