Our Motivation Masking Slide

Our Motivation • Naïve convolutions treat all input pixels as valid ones, thus are ill-fitted. • PartialConv¹ only tried to address this problem by binary masking on inputs. • However, binary masking is not ideal for internal feature maps. 1: Image Inpainting for Irregular Holes Using Partial Convolutions, Liu et al., ECCV, 2018

A conference presentation slide titled “Our Motivation” explains the limitations of naïve convolutions and binary masking, illustrated with a dark image of a person on horseback and a white masked silhouette. The joke is the unintentionally surreal visual example: the masking removes the central subject so aggressively that the technical explanation looks like an absurd reaction image. It is shared when a solution hides or erases the important part of something. EN: A conference presentation slide titled “Our Motivation” explains the limitations of naïve convolutions and binary masking, illustrated with a dark image of a person on horseback and a white masked silhouette. The joke is the unintentionally surreal visual example: the masking removes the central subject so aggressively that the technical explanation looks like an absurd reaction image. It is shared when a solution hides or erases the important part of something.

• @ithueti / 1573 2019-10-31