Fig. 4
From: Learning rich features with hybrid loss for brain tumor segmentation
The architecture of Multi-scale Feature Fusing Network (best view in color). 16×, 8×, 4×, and 2× represent the up sampling factors for feature maps of sizes 16 * 16, 32 * 32, 64 * 64, and 128 * 128, respectively. The Feature Fusing Block includes three convolution units, each composed of a convolution layer, a BN layer and an activation layer