Multiscale Efficient Channel Attention for Fusion Lane Line Segmentation

Figure 7

Loss curve during training; all training uses the pretraining parameters of ResNet34.