Fig. 2
From: Learning rich features with hybrid loss for brain tumor segmentation
The architecture of the proposed model (best view in color). Each block in Feature Extraction Network represents feature maps with different size (from left to right): 256*256, 128*128, 64*64, 32*32 and 16*16. In the multi-scale feature fusion network, the multi-scale feature maps are first up-sampled to the same size (176 × 176), and then fused