PFPN.py

        _, C2, C3, C4, C5 = resnet_graph(input_image, "resnet50", stage5=True)
        P5 = KL.Conv2D(256, (1, 1), name='fpn_c5p5')(C5)
        P4 = KL.Add(name="fpn_p4add")([
            KL.UpSampling2D(size=(2, 2), name="fpn_p5upsampled")(P5),
            KL.Conv2D(256, (1, 1), name='fpn_c4p4')(C4)])
        P3 = KL.Add(name="fpn_p3add")([
            KL.UpSampling2D(size=(2, 2), name="fpn_p4upsampled")(P4),
            KL.Conv2D(256, (1, 1), name='fpn_c3p3')(C3)])
        P2 = KL.Add(name="fpn_p2add")([
            KL.UpSampling2D(size=(2, 2), name="fpn_p3upsampled")(P3),
            KL.Conv2D(256, (1, 1), name='fpn_c2p2')(C2)])
        # New branch
        #Ni+1=Ni_down+P_(i+1)
        N2= P2
        N3 = KL.Add(name="fpn_p3n3add")([
            KL.Conv2D(256, (3, 3), strides=2, name="fpn_p2downsampled", padding="same")(N2),P3]) 
        N4 = KL.Add(name="fpn_p4n4add")([
            KL.Conv2D(256, (3, 3), strides=2, name="fpn_p3downsampled",padding="same")(N3), P4])  

        N5 = KL.Add(name="fpn_p5n5add")([
            KL.Conv2D(256, (3, 3), strides=2, name="fpn_p4downsampled",padding="same")(N4), P5]) 
        # Attach 3x3 conv to all P layers to get the final feature maps.
        N2 = KL.Conv2D(256, (3, 3), padding="SAME", name="fpn_n2")(N2)
        N3 = KL.Conv2D(256, (3, 3), padding="SAME", name="fpn_n3")(N3)
        N4 = KL.Conv2D(256, (3, 3), padding="SAME", name="fpn_n4")(N4)
        N5 = KL.Conv2D(256, (3, 3), padding="SAME", name="fpn_n5")(N5)
        # P6 is used for the 5th anchor scale in RPN. Generated by
        # subsampling from P5 with stride of 2.
        N6 = KL.MaxPooling2D(pool_size=(1, 1), strides=2, name="fpn_p6")(N5)

        rpn_feature_maps = [N2, N3, N4, N5, N6]
        mrcnn_feature_maps = [N2, N3, N4, N5]