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

Figure 6

Spatial Attention heatmaps from the Attention SPP and Attention RefineNet toward different-scale receptive fields. The Attention RefineNet is much closer to the output of model, and the Attention SPP contains much high-level abstract information. Therefore, the heatmaps of the Attention RefineNet have more detailed texture than those of the Attention SPP. (a) Input image. (b) Spatial heatmaps from Attention SPP. (c) Spatial heatmaps from Attention RefineNet.