Skip to main content
. 2020 Dec 18;12(1):494–508. doi: 10.1364/BOE.408471

Fig. 8.

Fig. 8.

Attention masks output by the deep learning model at different compression levels for 1) model trained on CRF = 12 and 2) model trained on matching CRF. As the compression increases, models trained on matching CRF focus more around the edges of the face, whereas models trained on CRF = 12 consistently focus on the skin regions.