RGBDOdometry.cpp

Go to the documentation of this file.

// ----------------------------------------------------------------------------
// -                        CloudViewer: www.cloudViewer.org                  -
// ----------------------------------------------------------------------------
// Copyright (c) 2018-2024 www.cloudViewer.org
// SPDX-License-Identifier: MIT
// ----------------------------------------------------------------------------
 
#include "t/pipelines/odometry/RGBDOdometry.h"
 
#include <benchmark/benchmark.h>
 
#include "camera/PinholeCameraIntrinsic.h"
#include "cloudViewer/core/CUDAUtils.h"
#include "cloudViewer/core/Tensor.h"
#include "cloudViewer/data/Dataset.h"
#include "cloudViewer/t/geometry/Image.h"
#include "cloudViewer/t/geometry/PointCloud.h"
#include "cloudViewer/t/io/ImageIO.h"
#include "cloudViewer/t/io/PointCloudIO.h"
 
namespace cloudViewer {
namespace t {
namespace pipelines {
namespace odometry {
 
static core::Tensor CreateIntrisicTensor() {
    camera::PinholeCameraIntrinsic intrinsic = camera::PinholeCameraIntrinsic(
            camera::PinholeCameraIntrinsicParameters::PrimeSenseDefault);
    auto focal_length = intrinsic.GetFocalLength();
    auto principal_point = intrinsic.GetPrincipalPoint();
    return core::Tensor::Init<double>(
            {{(focal_length.first), 0, (principal_point.first)},
             {0, (focal_length.second), (principal_point.second)},
             {0, 0, 1}});
}
 
static void ComputeOdometryResultPointToPlane(benchmark::State& state,
                                              const core::Device& device) {
    if (!t::geometry::Image::HAVE_IPP && device.IsCPU()) {
        return;
    }
 
    const float depth_scale = 1000.0;
    const float depth_diff = 0.07;
    const float depth_max = 3.0;
 
    data::SampleRedwoodRGBDImages redwood_data;
    t::geometry::Image src_depth =
            *t::io::CreateImageFromFile(redwood_data.GetDepthPaths()[0]);
    t::geometry::Image dst_depth =
            *t::io::CreateImageFromFile(redwood_data.GetDepthPaths()[2]);
 
    src_depth = src_depth.To(device);
    dst_depth = dst_depth.To(device);
 
    core::Tensor intrinsic_t = CreateIntrisicTensor();
    t::geometry::Image src_depth_processed =
            src_depth.ClipTransform(depth_scale, 0.0, depth_max, NAN);
    t::geometry::Image src_vertex_map =
            src_depth_processed.CreateVertexMap(intrinsic_t, NAN);
    t::geometry::Image src_normal_map = src_vertex_map.CreateNormalMap(NAN);
 
    t::geometry::Image dst_depth_processed =
            dst_depth.ClipTransform(depth_scale, 0.0, depth_max, NAN);
    t::geometry::Image dst_vertex_map =
            dst_depth_processed.CreateVertexMap(intrinsic_t, NAN);
 
    core::Tensor trans =
            core::Tensor::Eye(4, core::Float64, core::Device("CPU:0"));
 
    for (int i = 0; i < 20; ++i) {
        auto result = t::pipelines::odometry::ComputeOdometryResultPointToPlane(
                src_vertex_map.AsTensor(), dst_vertex_map.AsTensor(),
                src_normal_map.AsTensor(), intrinsic_t, trans, depth_diff,
                depth_diff * 0.5);
        trans = result.transformation_.Matmul(trans).Contiguous();
    }
 
    for (auto _ : state) {
        core::Tensor trans =
                core::Tensor::Eye(4, core::Float64, core::Device("CPU:0"));
 
        for (int i = 0; i < 20; ++i) {
            auto result =
                    t::pipelines::odometry::ComputeOdometryResultPointToPlane(
                            src_vertex_map.AsTensor(),
                            dst_vertex_map.AsTensor(),
                            src_normal_map.AsTensor(), intrinsic_t, trans,
                            depth_diff, depth_diff * 0.5);
            trans = result.transformation_.Matmul(trans).Contiguous();
        }
        core::cuda::Synchronize(device);
    }
}
 
static void RGBDOdometryMultiScale(
        benchmark::State& state,
        const core::Device& device,
        const t::pipelines::odometry::Method& method) {
    if (!t::geometry::Image::HAVE_IPP && device.IsCPU()) {
        return;
    }
 
    const float depth_scale = 1000.0;
    const float depth_max = 3.0;
    const float depth_diff = 0.07;
 
    data::SampleRedwoodRGBDImages redwood_data;
    t::geometry::Image src_depth =
            *t::io::CreateImageFromFile(redwood_data.GetDepthPaths()[0]);
    t::geometry::Image src_color =
            *t::io::CreateImageFromFile(redwood_data.GetColorPaths()[0]);
 
    t::geometry::Image dst_depth =
            *t::io::CreateImageFromFile(redwood_data.GetDepthPaths()[2]);
 
    t::geometry::Image dst_color =
            *t::io::CreateImageFromFile(redwood_data.GetColorPaths()[0]);
 
    t::geometry::RGBDImage source, target;
    source.color_ = src_color.To(device);
    source.depth_ = src_depth.To(device);
    target.color_ = dst_color.To(device);
    target.depth_ = dst_depth.To(device);
 
    core::Tensor intrinsic_t = CreateIntrisicTensor();
 
    // Very strict criteria to ensure running most the iterations
    t::pipelines::odometry::OdometryLossParams loss(depth_diff);
    std::vector<t::pipelines::odometry::OdometryConvergenceCriteria> criteria{
            t::pipelines::odometry::OdometryConvergenceCriteria(10, 1e-12,
                                                                1e-12),
            t::pipelines::odometry::OdometryConvergenceCriteria(5, 1e-12,
                                                                1e-12),
            t::pipelines::odometry::OdometryConvergenceCriteria(3, 1e-12,
                                                                1e-12)};
 
    // Warp up
    RGBDOdometryMultiScale(
            source, target, intrinsic_t,
            core::Tensor::Eye(4, core::Float64, core::Device("CPU:0")),
            depth_scale, depth_max, criteria, method, loss);
 
    for (auto _ : state) {
        RGBDOdometryMultiScale(
                source, target, intrinsic_t,
                core::Tensor::Eye(4, core::Float64, core::Device("CPU:0")),
                depth_scale, depth_max, criteria, method, loss);
        core::cuda::Synchronize(device);
    }
}
 
BENCHMARK_CAPTURE(ComputeOdometryResultPointToPlane, CPU, core::Device("CPU:0"))
        ->Unit(benchmark::kMillisecond);
#ifdef BUILD_CUDA_MODULE
BENCHMARK_CAPTURE(ComputeOdometryResultPointToPlane,
                  CUDA,
                  core::Device("CUDA:0"))
        ->Unit(benchmark::kMillisecond);
#endif
 
BENCHMARK_CAPTURE(RGBDOdometryMultiScale,
                  Hybrid_CPU,
                  core::Device("CPU:0"),
                  t::pipelines::odometry::Method::Hybrid)
        ->Unit(benchmark::kMillisecond);
BENCHMARK_CAPTURE(RGBDOdometryMultiScale,
                  Intensity_CPU,
                  core::Device("CPU:0"),
                  t::pipelines::odometry::Method::Intensity)
        ->Unit(benchmark::kMillisecond);
BENCHMARK_CAPTURE(RGBDOdometryMultiScale,
                  PointToPlane_CPU,
                  core::Device("CPU:0"),
                  t::pipelines::odometry::Method::PointToPlane)
        ->Unit(benchmark::kMillisecond);
 
#ifdef BUILD_CUDA_MODULE
BENCHMARK_CAPTURE(RGBDOdometryMultiScale,
                  Hybrid_CUDA,
                  core::Device("CUDA:0"),
                  t::pipelines::odometry::Method::Hybrid)
        ->Unit(benchmark::kMillisecond);
BENCHMARK_CAPTURE(RGBDOdometryMultiScale,
                  Intensity_CUDA,
                  core::Device("CUDA:0"),
                  t::pipelines::odometry::Method::Intensity)
        ->Unit(benchmark::kMillisecond);
BENCHMARK_CAPTURE(RGBDOdometryMultiScale,
                  PointToPlane_CUDA,
                  core::Device("CUDA:0"),
                  t::pipelines::odometry::Method::PointToPlane)
        ->Unit(benchmark::kMillisecond);
#endif
}  // namespace odometry
}  // namespace pipelines
}  // namespace t
}  // namespace cloudViewer