Compilation issues with PCL and message_filters: ; cannot bind non-const lvalue

edit

Compiler: gcc9.4.0 Arch: Ubuntu20.04/AMD64 PCL: 1.10 ROS Galactic

I've got two errors which I believe are hindering me from compiling my code here successfully. The first is affecting my attempt to calculate the 3D centroid using PCL:

cannot bind non-const lvalue reference of type ‘Eigen::Matrix<int, 4, 1>&’ to an rvalue of type ‘Eigen::Matrix<int, 4, 1>’

I've tried a few different type, including pointers/non-pointers, but I'm a bit stumped with this.

The second is an issue with using message_filters effectively. I've tried to copy the approach used by image_pipeline for stereography, but it's throwing a number of errors:

error: no matching function for call to ‘message_filters::MessageEvent<const vision_msgs::msg::Detection2D_<std::allocator<void> > >::MessageEvent(const message_filters::MessageEvent<const sensor_msgs::msg::PointCloud2_<std::allocator<void> > >&, bool)’
error: no matching function for call to ‘message_filters::Signal9<vision_msgs::msg::Detection2D_<std::allocator<void> >, sensor_msgs::msg::PointCloud2_<std::allocator<void> >, message_filters::NullType, message_filters::NullType, message_filters::NullType, message_filters::NullType, message_filters::NullType, message_filters::NullType, message_filters::NullType>::addCallback<const M0ConstPtr&, const M1ConstPtr&, const M2ConstPtr&, const M3ConstPtr&, const M4ConstPtr&, const M5ConstPtr&, const M6ConstPtr&, const M7ConstPtr&, const M8ConstPtr&>(std::_Bind_helper<false, const std::_Bind<void (pose_eval::PoseEval::*(pose_eval::PoseEval*, std::_Placeholder<1>, std::_Placeholder<2>))(sensor_msgs::msg::PointCloud2_<std::allocator<void> >&, vision_msgs::msg::Detection2D_<std::allocator<void> >&)>&, const std::_Placeholder<1>&, const std::_Placeholder<2>&, const std::_Placeholder<3>&, const std::_Placeholder<4>&, const std::_Placeholder<5>&, const std::_Placeholder<6>&, const std::_Placeholder<7>&, const std::_Placeholder<8>&, const std::_Placeholder<9>&>::type)’
[5.100s]   280 | ::_3, std::placeholders::_4, std::placeholders::_5, std::placeholders::_6, std::placeholders::_7, std::placeholders::_8, std::placeholders::_9));

The latter error might be causing the former, but I'm not 100% on that. Full error log is not attached, so please find it at the bottom here.

The relevant code is follows:

PoseEval::PoseEval(rclcpp::NodeOptions options) : Node("pose_eval_cpp", options) {
    rclcpp::QoS qos = rclcpp::SystemDefaultsQoS();
    std::string ns = std::string(this->get_namespace());
    this->detect_loc_pub_ = this->create_publisher<PointStamped>(ns + std::string("/detect_pos"), qos);

    message_filters::Subscriber<PointCloud2> pc2_sub_(this, ns + "", qos.get_rmw_qos_profile());
    message_filters::Subscriber<Detection2D> best_detect_sub_(this, ns + "", qos.get_rmw_qos_profile());

    this->synch_subs_.reset(new ApproximateSync(ApproximatePolicy(20), pc2_sub_, best_detect_sub_));
    this->synch_subs_->registerCallback(std::bind(&PoseEval::points_callback, this, std::placeholders::_1, std::placeholders::_2));
}

void PoseEval::points_callback(PointCloud2& points, Detection2D& best_match) {
    RCLCPP_INFO(this->get_logger(), "Synching detection and cloud");

    // Find best result
    auto byScore = [&](const ObjectHypothesisWithPose& a, const ObjectHypothesisWithPose& b) {
        return a.hypothesis.score < b.hypothesis.score;
    };
    auto best = std::max_element(best_match.results.begin(), best_match.results.end(), byScore);

    auto best_class = best->hypothesis.class_id;
    auto best_score = best->hypothesis.score;

    // Get ROI points
    auto box = best_match.bbox;
    auto box_c = box.center;
    auto top = box_c.y - box.size_y / 2, bottom = box_c.y + box.size_y / 2, left = box_c.x - box.size_x / 2, right = box_c.x + box.size_x / 2;

    pcl::PointCloud<pcl::PointXYZRGB>::Ptr cloud = boost::make_shared<pcl::PointCloud<pcl::PointXYZRGB>>();
    pcl::PointCloud<pcl::PointXYZRGB>::Ptr cloud_subset = boost::make_shared<pcl::PointCloud<pcl::PointXYZRGB>>();

    pcl::fromROSMsg(points, *cloud);

    // Extract ROI
    pcl::ExtractIndices<pcl::PointXYZRGB> filter(true);
    filter.setInputCloud(cloud);
    filter.setIndices(top, left, bottom - top, right - left);
    filter.filter(*cloud_subset);

    // Find Centroid
    auto centroid = Eigen::Vector4f(0, 0, 0, 0);
    auto n_pts_used = pcl::compute3DCentroid<pcl::PointXYZRGB ...