61 lines
1.7 KiB
C++
Executable File
61 lines
1.7 KiB
C++
Executable File
// Copyright 2016, Tobias Hermann.
|
|
// https://github.com/Dobiasd/frugally-deep
|
|
// Distributed under the MIT License.
|
|
// (See accompanying LICENSE file or at
|
|
// https://opensource.org/licenses/MIT)
|
|
|
|
#pragma once
|
|
|
|
#include "fdeep/layers/layer.hpp"
|
|
|
|
#include <fplus/fplus.hpp>
|
|
|
|
#include <cassert>
|
|
#include <cstddef>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
namespace fdeep { namespace internal
|
|
{
|
|
|
|
class upsampling_2d_layer : public layer
|
|
{
|
|
public:
|
|
explicit upsampling_2d_layer(const std::string& name,
|
|
const shape2& scale_factor) :
|
|
layer(name),
|
|
scale_factor_(scale_factor)
|
|
{
|
|
}
|
|
protected:
|
|
tensor5s apply_impl(const tensor5s& inputs) const override final
|
|
{
|
|
assertion(inputs.size() == 1, "invalid number of inputs tensors");
|
|
const auto& input = inputs.front();
|
|
return {upsampling2d(input)};
|
|
}
|
|
shape2 scale_factor_;
|
|
tensor5 upsampling2d(const tensor5& in_vol) const
|
|
{
|
|
tensor5 out_vol(shape5(1, 1,
|
|
in_vol.shape().height_ * scale_factor_.height_,
|
|
in_vol.shape().width_ * scale_factor_.width_,
|
|
in_vol.shape().depth_), 0);
|
|
for (std::size_t z = 0; z < in_vol.shape().depth_; ++z)
|
|
{
|
|
for (std::size_t y = 0; y < out_vol.shape().height_; ++y)
|
|
{
|
|
std::size_t y_in = y / scale_factor_.height_;
|
|
for (std::size_t x = 0; x < out_vol.shape().width_; ++x)
|
|
{
|
|
std::size_t x_in = x / scale_factor_.width_;
|
|
out_vol.set(0, 0, y, x, z, in_vol.get(0, 0, y_in, x_in, z));
|
|
}
|
|
}
|
|
}
|
|
return out_vol;
|
|
}
|
|
};
|
|
|
|
} } // namespace fdeep, namespace internal
|