torch::nn::Sequential c2,

torch::nn::Sequential c3,

torch::nn::Sequential c4,

torch::nn::Sequential c5,

uint32_t out_channels)

: c1_(c1),

c2_(c2),

c3_(c3),

c4_(c4),

c5_(c5),

p6_(torch::nn::Functional(torch::max_pool2d,

/*kernel_size*/ at::IntList({1}),

/*stride*/ at::IntList({2}),

/*padding*/ at::IntList({0}),

/*dilation*/ at::IntList({1}),

/*ceil_mode*/ false)),

p5_conv1_(torch::nn::Conv2dOptions(2048, out_channels, 1).stride(1)),

p5_conv2_(SamePad2d(/*kernel_size*/ 3, /*stride*/ 1),

torch::nn::Conv2d(

torch::nn::Conv2dOptions(out_channels, out_channels, 3)

.stride(1))),

p4_conv1_(torch::nn::Conv2dOptions(1024, out_channels, 1).stride(1)),

p4_conv2_(SamePad2d(/*kernel_size*/ 3, /*stride*/ 1),

torch::nn::Conv2d(

torch::nn::Conv2dOptions(out_channels, out_channels, 3)

.stride(1))),

p3_conv1_(torch::nn::Conv2dOptions(512, out_channels, 1).stride(1)),

p3_conv2_(SamePad2d(/*kernel_size*/ 3, /*stride*/ 1),

torch::nn::Conv2d(

torch::nn::Conv2dOptions(out_channels, out_channels, 3)

.stride(1))),

p2_conv1_(torch::nn::Conv2dOptions(256, out_channels, 1).stride(1)),

p2_conv2_(SamePad2d(/*kernel_size*/ 3, /*stride*/ 1),

torch::nn::Conv2d(

torch::nn::Conv2dOptions(out_channels, out_channels, 3)

.stride(1))) {

register_module("C1", c1_);

register_module("C2", c2_);

register_module("C3", c3_);

register_module("C4", c4_);

register_module("C5", c5_);

register_module("P6", p6_);

register_module("P5_conv1", p5_conv1_);

register_module("P5_conv2", p5_conv2_);

register_module("P4_conv1", p4_conv1_);

register_module("P4_conv2", p4_conv2_);

register_module("P3_conv1", p3_conv1_);

register_module("P3_conv2", p3_conv2_);

register_module("P2_conv1", p2_conv1_);

register_module("P2_conv2", p2_conv2_);

x = c1_->forward(x);

x = c2_->forward(x);

auto c2_out = x;

x = c3_->forward(x);

auto c3_out = x;

x = c4_->forward(x);

auto c4_out = x;

x = c5_->forward(x);

auto p5_out = p5_conv1_->forward(x);

auto p4_out =

p4_conv1_->forward(c4_out) + upsample(p5_out, /*scale_factor*/ 2);

auto p3_out =

p3_conv1_->forward(c3_out) + upsample(p4_out, /*scale_factor*/ 2);

auto p2_out =

p2_conv1_->forward(c2_out) + upsample(p3_out, /*scale_factor*/ 2);

p5_out = p5_conv2_->forward(p5_out);

p4_out = p4_conv2_->forward(p4_out);

p3_out = p3_conv2_->forward(p3_out);

p2_out = p2_conv2_->forward(p2_out);

// P6 is used for the 5th anchor scale in RPN. Generated by subsampling from

// P5 with stride of 2.

auto p6_out = p6_->forward(p5_out);

return {p2_out, p3_out, p4_out, p5_out, p6_out};