Split操作:使用多个卷积核进行卷积,形成多个分支。这里的 Spli

Split operation: Use multiple convolution kernels for convolution to form multiple branches. Split here is a complete convolution operation of different convolution kernel sizes on the input vector X. Two branches of 3×3 and 5×5 convolution kernels can be used. In order to further improve the efficiency and save the amount of calculation, the conventional 5x5 convolution can be replaced with a 3x3 hole convolution. The hole convolution has a better effect in image segmentation. The hole convolution is shown in Figure 3.9:

Split operation: Reels are made using multiple co product cores to form multiple branches. Split here is a full reel operation with different co product sizes for the input vector X. You can × two branches of × 3, 3 and 5 and 5. In order to further improve the efficiency and save the calculation, the conventional 5x5 reel can be replaced with 3x3 hollow reel, which has a good effect in image segmentation. The hollow reel is shown in Figure 3.9:

Split operation: convolution with multiple convolution kernels to form multiple branches. Here, split is a complete convolution operation on the input vector x with different convolution kernel sizes. Two branches of convolution kernel of 3 × 3 and 5 × 5 can be used. In order to further improve the efficiency and save the amount of computation, the conventional 5x5 convolution can be replaced by 3x3 hole convolution, which has a good effect in image segmentation. The hole convolution is shown in Figure 3.9<br>