Semantic segmentation model with the fusion of deeplabv3plus, hed and attention loss.
Rui Guo, Ronghua Liu, Na Li, Wei Liu
DV3+HED+: A DCNNs-based Framework to Monitor Temporary Works and ESAs in Railway Construction Project Using VHR Satellite Images
https://arxiv.org/abs/1908.11080
References:
[1]Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
[2]Holistically-Nested Edge Detection
[3]DOOBNet: Deep Object Occlusion Boundary Detection from an Image