SPP - Githubissues

代码中spp是这样的：

def Spp_Conv2D_BN_Leaky(x, num_filters):
    y1 = MaxPooling2D(pool_size=(5,5), strides=(1,1), padding='same')(x)
    y2 = MaxPooling2D(pool_size=(9,9), strides=(1,1), padding='same')(x)
    y3 = MaxPooling2D(pool_size=(13,13), strides=(1,1), padding='same')(x)

    y = compose(
            Concatenate(),
            DarknetConv2D_BN_Leaky(num_filters, (1,1)))([y1, y2, y3, x])
    return y

但在AlexeyAB/darknet/cfg/yolov4.cfg对SPP结构的描述为

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

### SPP ###
[maxpool]
stride=1
size=5

[route]
layers=-2

[maxpool]
stride=1
size=9

[route]
layers=-4

[maxpool]
stride=1
size=13

[route]
layers=-1,-3,-5,-6
### End SPP ###

可以看出实际SPP应该是DarknetConv2D_BN_Leaky(num_filters, (1,1)))([ y3, y2, y1, x]) 顺序问题已经对检测结果造成了影响

david8862 / keras-YOLOv3-model-set

SPP #132