Introducing HANet: An Add-On Module For Urban-Scene Segmentation

HANet (height-driven attention networks) is an add-on module introduced by this research group via a paper to exploit the intrinsic features of urban-scene images. HANet can be used for improving semantic segmentation for urban-scene images. The module focusses on informative features or classes selectively according to the vertical position of a pixel.

While most semantic segmentation networks do not consider unique attributes in urban-scene images that have their own distinct characteristics, HANet incorporates the capability exploiting the attributes to handle the urban scene dataset effectively. According to the paper, the researchers mention that they have validated the consistent performance (mIoU) increase of various semantic segmentation models on two datasets when HANet is adopted. This brings to this conclusion that by adding this module (HANet) to existing models will be cost-effective and easy.