Multiscale Meets Spatial Awareness: An Efficient Attention Guidance Network for Human Parsing

Figure 1

Illustration of challenging scales and spatial variations of real-world human images from the LIP dataset [3]. In (a), the scales of the upper-body and the head are larger than that of the lower-body. In (b), due to the transverse distribution of a body, the spatial locations of human parts are different from usual human images. In (c), both scale and spatial variations appear.

(a)

(b)

(c)