pyvale 2025.5.3__cp311-cp311-musllinux_1_2_aarch64.whl → 2025.7.1__cp311-cp311-musllinux_1_2_aarch64.whl

pyvale/dic/cpp/dicscanmethod.cpp

@@ -0,0 +1,677 @@
+// ================================================================================
+// pyvale: the python validation engine
+// License: MIT
+// Copyright (C) 2025 The Computer Aided Validation Team
+// ================================================================================
+
+
+// STD library Header files
+#include <queue>
+#include <atomic>
+#include <thread>
+#include <cstring>
+#include <omp.h>
+#include <csignal>
+
+// Program Header files
+#include "./dicbruteforce.hpp"
+#include "./dicinterpolator.hpp"
+#include "./dicoptimizer.hpp"
+#include "./defines.hpp"
+#include "./dicutil.hpp"
+#include "./dicrg.hpp"
+#include "./indicators.hpp"
+#include "./cursor_control.hpp"
+#include "./dicfourier.hpp"
+
+namespace scanmethod {
+
+
+    // for graceful exit
+    std::atomic<bool> stop_request(false);
+
+    void signalHandler(int signal) {
+        if (signal == SIGINT) {
+            stop_request = true;
+        }
+    }
+
+    void image(const double *img_ref,
+               const Interpolator &interp_def,
+               const util::SubsetData &ssdata, 
+               const util::Config &conf,
+               const int img_num){
+
+        const int num_ss = ssdata.num;
+        const int ss_size = ssdata.size;
+
+        // progress bar
+        indicators::ProgressBar bar;
+        util::create_progress_bar(bar, conf.filenames, img_num, 100);
+        std::atomic<int> current_progress = 0;
+        std::atomic<int> prev_pct = 0;
+
+               // loop over subsets within the ROI
+        #pragma omp parallel shared(stop_request)
+        {
+
+            // initialise subsets
+            util::Subset ss_def(ss_size);
+            util::Subset ss_ref(ss_size);
+
+            // optimization parameters
+            optimizer::Parameters opt(conf.num_params, conf.max_iter,
+                                    conf.precision, conf.opt_threshold,
+                                    conf.px_vert, conf.px_hori);
+
+            // if using SSD then not going to use opt_threshold. It can take
+            // any value. Convergence will be checked against precision only
+            if (conf.corr_crit=="SSD")
+                opt.opt_threshold = std::numeric_limits<double>::max();
+
+
+            #pragma omp for
+            for (int ss = 0; ss < num_ss; ss++){
+
+                // exit the main DIC loop when ctrl+C is hit
+                if (stop_request){
+                    continue;
+                }
+
+                // subset coordinate list takes central locations. 
+                // Converting to top left corner for optimization routine
+                int ss_x = ssdata.coords[ss*2];
+                int ss_y = ssdata.coords[ss*2+1];
+
+                // get the reference subset
+                util::extract_ss(ss_ref, ss_x, ss_y, conf.px_hori, conf.px_vert, img_ref);
+
+                for (int i = 0; i < opt.num_params; i++){
+                    opt.p[i] = 0.0;
+                }
+
+                // perform optimization on subset from deformed image
+                double centre_x = ss_x + static_cast<double>(ssdata.size)/2.0 - 0.5;
+                double centre_y = ss_y + static_cast<double>(ssdata.size)/2.0 - 0.5;
+                util::Results res = optimizer::solve(centre_x, centre_y, ss_ref, ss_def, interp_def, opt, conf.corr_crit);
+
+                if (conf.corr_crit!="SSD")
+                    res.cost = 1-res.cost;
+
+                // append the results for the current subset to result vectors
+                util::append_results(img_num, ss, res, num_ss);
+
+                // update progress bar
+                int progress = current_progress.fetch_add(1);
+                util::update_progress_bar(bar, progress, num_ss, prev_pct);
+
+            }
+        }
+        bar.mark_as_completed();
+        indicators::show_console_cursor(true);
+
+    }
+
+
+
+    void image_with_bf(const double *img_ref,
+                    const double *img_def,
+                    const Interpolator &interp_def,
+                    const util::SubsetData  &ssdata, 
+                    const util::Config &conf,
+                    const int img_num){
+
+        const int num_ss = ssdata.num;
+        const int ss_size = ssdata.size;
+
+        // progress bar
+        indicators::ProgressBar bar;
+        util::create_progress_bar(bar, conf.filenames, img_num, num_ss);
+        std::atomic<int> current_progress = 0;
+        std::atomic<int> prev_pct = 0;
+
+        // initialise subsets
+        util::Subset ss_def(ss_size);
+        util::Subset ss_ref(ss_size);
+
+        // optimization parameters
+        optimizer::Parameters opt(conf.num_params, conf.max_iter, 
+                                conf.precision, conf.opt_threshold,
+                                conf.px_vert, conf.px_hori);
+
+        // if using SSD then not going to use opt_threshold. It can take
+        // any value. Convergence will be checked against precision only
+        if (conf.corr_crit=="SSD")
+            opt.opt_threshold = std::numeric_limits<double>::max();
+
+
+        // brute force scan parameters
+        brute::Parameters brute(conf.bf_threshold, conf.max_disp);
+
+        // perform optimization on subset from deformed image
+        util::Results res(conf.num_params);
+
+        // counter for each thread
+        int ss_thread_num = 0;
+
+        // temp p values for copy from brute force to optimization.
+        double ptemp[6] = {0,0,0,0,0,0};
+
+        // loop over subsets within the ROI
+        #pragma omp parallel for firstprivate(ss_ref, ss_def, ss_thread_num, opt, brute, res, ptemp)
+        for (int ss = 0; ss < num_ss; ss++){
+
+            // exit the main DIC loop when ctrl+C is hit
+            if (stop_request){
+                continue;
+            }
+
+
+            // subset coordinate list contains central locations.
+            // Converting to top left corner for optimization routine
+            int ss_x = ssdata.coords[ss*2];
+            int ss_y = ssdata.coords[ss*2+1];
+
+            // get the reference subset values from the reference img
+            util::extract_ss(ss_ref, ss_x, ss_y, conf.px_hori, conf.px_vert, img_ref); 
+
+
+            // if first subset in the loop or prev subset was a poor match
+            // start search with a brute force scan using the last set of 
+            // brute force params that gave a good match.
+            if ((ss_thread_num == 0) || (res.cost > opt.opt_threshold)){
+
+                brute::expanding_wavefront(ss_x, ss_y, img_ref, 
+                                        conf.px_hori, 
+                                        conf.px_vert, 
+                                        ss_ref, ss_def, brute);
+
+                ptemp[0] = brute.p_rigid[0];
+                ptemp[1] = brute.p_rigid[1];
+
+                for (int i = 0; i < opt.num_params; i++){
+                    opt.p[i] = ptemp[i];
+                }
+            }
+
+            double centre_x = ss_x + static_cast<double>(ssdata.size)/2.0 - 0.5;
+            double centre_y = ss_y + static_cast<double>(ssdata.size)/2.0 - 0.5;
+            util::Results res = optimizer::solve(centre_x, centre_y, ss_ref, ss_def, interp_def, opt, conf.corr_crit);
+
+
+            // if its not SSD, then we need to flip the cost values so that 1.0
+            // is a perfect match rather than 0.0
+            if (conf.corr_crit!="SSD")
+                res.cost = 1-res.cost;
+
+            // append the results for the current subset to result vectors
+            util::append_results(img_num, ss, res, num_ss);
+
+            ss_thread_num++;
+
+            // update progress bar
+            int progress = current_progress.fetch_add(1);
+            util::update_progress_bar(bar, progress, num_ss, prev_pct);
+
+        }
+        bar.mark_as_completed();
+        indicators::show_console_cursor(true);
+    }
+
+
+
+
+
+
+    void reliability_guided(const double *img_ref,
+                            const double *img_def,
+                            const Interpolator &interp_def,
+                            const std::vector<util::SubsetData> &ssdata,
+                            const util::Config &conf,
+                            const int img_num,
+                            const bool save_at_end){
+
+        // assign some consts for readability
+        const int px_hori = conf.px_hori;
+        const int px_vert = conf.px_vert;
+        const int seed_x = conf.rg_seed.first;
+        const int seed_y = conf.rg_seed.second;
+        const int nsizes = ssdata.size();
+        const int last_size = nsizes-1;
+        const int num_ss = ssdata[last_size].num;
+        const int ss_size = ssdata[last_size].size;
+        const int ss_step = ssdata[last_size].step;
+
+
+        fourier::mgwd(ssdata, img_ref, img_def, interp_def, 
+                      conf.fft_mad, conf.fft_mad_scale);
+
+        // progress bar
+        indicators::ProgressBar bar;
+        util::create_progress_bar(bar, conf.filenames, img_num, num_ss);
+        std::atomic<int> current_progress(0);
+        std::atomic<int> prev_pct(0);
+
+        // quick check for the initial seed point
+        if (!rg::is_valid_point(seed_x, seed_y, ssdata[last_size])) {
+            return;
+        }
+
+        // Initialize binary mask for computed points (initialized to 0)
+        std::vector<std::atomic<int>> computed_mask(ssdata[last_size].mask.size());
+        for (auto& val : computed_mask) val.store(0);
+
+        // queue for each thread
+        std::vector<std::priority_queue<rg::Point>> local_q(omp_get_max_threads());
+
+        // Mutex vector to protect each queue
+        std::vector<std::mutex> queue_mutexes(omp_get_max_threads());
+
+        # pragma omp parallel
+        {
+
+            int tid = omp_get_thread_num();
+            std::priority_queue<rg::Point>& thread_q = local_q[tid];
+
+            // Initialize ref and def subsets
+            util::Subset ss_def(ss_size);
+            util::Subset ss_ref(ss_size);
+
+            // Optimization parameters
+            optimizer::Parameters opt(conf.num_params, conf.max_iter, 
+                                      conf.precision, conf.opt_threshold, 
+                                      px_vert, px_hori);
+
+            // if using SSD then not going to use opt_threshold. It can take
+            // any value. Convergence will be checked against precision only
+            if (conf.corr_crit=="SSD")
+                opt.opt_threshold = std::numeric_limits<double>::max();
+
+            brute::Parameters brute(conf.bf_threshold, conf.max_disp);
+
+            std::vector<std::unique_ptr<fourier::FFT>> fft_windows;
+
+            for (size_t t = 0; t < ssdata.size(); ++t) {
+                fft_windows.push_back(std::make_unique<fourier::FFT>(ssdata[t].size));
+            }
+
+            // TODO: for the seed location I'm going to overwride the max 
+            // number of iterations to make sure we get a good convergence.
+            // this is hardcoded for now. Could do with updating so that 
+            // the seed location is checked ahead of the main correlation run.
+            opt.max_iter = 200;
+
+            // ---------------------------------------------------------------------------------------------------------------------------
+            // PROCESS THE SEED SUBSET 
+            // ---------------------------------------------------------------------------------------------------------------------------
+            if (tid == 0) {
+
+                // seed coordinates
+                int x = seed_x / ss_step;
+                int y = seed_y / ss_step;
+                int idx = ssdata[last_size].mask[y * ssdata[last_size].num_ss_x + x];
+
+
+                // if the first image. Take the optimization parameters from rigid fourier
+                std::fill(opt.p.begin(), opt.p.end(), 0.0);
+                opt.p[0] = fourier::shifts[last_size].x[idx];
+                opt.p[1] = fourier::shifts[last_size].y[idx];
+
+                // Extract reference subset and solve for starting seed point
+                util::extract_ss(ss_ref, seed_x, seed_y, px_hori, px_vert, img_ref);
+
+
+                double centre_x = seed_x + static_cast<double>(ss_size)/2.0 - 0.5;
+                double centre_y = seed_y + static_cast<double>(ss_size)/2.0 - 0.5;
+
+                util::Results seed_res = optimizer::solve(centre_x, centre_y, ss_ref, ss_def, interp_def, opt, conf.corr_crit);
+
+                // if its not SSD, then we need to flip the cost values so that 1.0
+                // is a perfect match rather than 0.0
+                if (conf.corr_crit!="SSD")
+                    seed_res.cost = 1.0-seed_res.cost;
+
+                // append the results for the current subset to result vectors
+                util::append_results(img_num, idx, seed_res, num_ss);
+
+                computed_mask[idx].store(1);
+
+                // int progress = current_progress.fetch_add(1);
+                // util::update_progress_bar(bar, progress, num_ss, prev_pct);
+
+                // loop over the neighbours for the initial seed point
+                for (size_t n = 0; n < ssdata[last_size].neigh[idx].size(); n++) {
+
+                    // subset index of neighbour to the current point
+                    int nidx = ssdata[last_size].neigh[idx][n];
+
+                    int nx = ssdata[last_size].coords[nidx*2];
+                    int ny = ssdata[last_size].coords[nidx*2+1];
+
+                    util::extract_ss(ss_ref, nx, ny, px_hori, px_vert, img_ref);
+
+                    // get parameter values from fft output or from previous image
+                    std::fill(opt.p.begin(), opt.p.end(), 0.0);
+                    opt.p[0] = fourier::shifts[last_size].x[nidx];
+                    opt.p[1] = fourier::shifts[last_size].y[nidx];
+
+                    // perform optimization for seed point neighbours
+                    double centre_x = nx + static_cast<double>(ss_size)/2.0 - 0.5;
+                    double centre_y = ny + static_cast<double>(ss_size)/2.0 - 0.5;
+                    util::Results nres = optimizer::solve(centre_x, centre_y, ss_ref, ss_def, interp_def, opt, conf.corr_crit);
+                    
+                    // if its not SSD, then we need to flip the cost values so that 1.0
+                    // is a perfect match rather than 0.0
+                    if (conf.corr_crit!="SSD")
+                        nres.cost = 1.0-nres.cost;
+
+                    // append the results for the current subset to result vectors
+                    util::append_results(img_num, nidx, nres, num_ss);
+
+                    // update mask
+                    computed_mask[idx].store(1);
+
+                    // add this point to queue
+                    // Protect push with mutex
+                    {
+                        std::lock_guard<std::mutex> lock(queue_mutexes[0]);
+                        local_q[0].push(rg::Point(nidx,nres.cost));
+                    }
+
+                    // update progress bar
+                    // int progress = current_progress.fetch_add(1);
+                    // util::update_progress_bar(bar, progress, num_ss, prev_pct);
+                }
+            }
+
+
+            // ---------------------------------------------------------------------------------------------------------------------------
+            // PROCESS ALL OTHER SUBSETS
+            // ---------------------------------------------------------------------------------------------------------------------------
+            #pragma omp barrier
+
+            opt.max_iter = conf.max_iter;
+
+            std::vector<rg::Point> temp_neigh;
+            temp_neigh.reserve(4);
+
+            const int max_idle_iters = 100;
+            rg::Point current(0, 0);
+
+            while (!stop_request) {
+                bool got_point = false;
+                int idle_iters = 0;
+
+                // Try own queue safely
+                {
+                    std::lock_guard<std::mutex> lock(queue_mutexes[tid]);
+                    if (!thread_q.empty()) {
+                        current = thread_q.top();
+                        thread_q.pop();
+                        got_point = true;
+                    }
+                }
+
+                // Steal if nothing in own queue
+                if (!got_point) {
+                    while (!got_point && idle_iters < max_idle_iters) {
+                        #pragma omp critical
+                        {
+                            for (size_t i = 0; i < local_q.size(); ++i) {
+                                std::lock_guard<std::mutex> lock(queue_mutexes[i]);
+                                if (!local_q[i].empty()) {
+                                    current = local_q[i].top();
+                                    local_q[i].pop();
+                                    got_point = true;
+                                    break;
+                                }
+                            }
+                        }
+                        if (!got_point) {
+                            ++idle_iters;
+                            std::this_thread::sleep_for(std::chrono::milliseconds(1));
+                        }
+                    }
+                }
+
+                if (!got_point) {
+                    break;
+                }
+
+                temp_neigh.clear();
+
+
+                // index of current point in results arrays
+                int idx_results = save_at_end ? img_num * num_ss + current.idx : current.idx;
+                int idx_results_p = idx_results * opt.num_params;
+
+                // loop over neighbouring points
+                for (size_t n = 0; n < ssdata[last_size].neigh[current.idx].size(); n++) {
+
+                    // subset index of neighbour to the current point
+                    int nidx = ssdata[last_size].neigh[current.idx][n];
+
+                    int expected = 0;
+                    expected = computed_mask[nidx].exchange(1);
+                    if (expected == 0) {
+
+                        // coords of neigh
+                        int nx = ssdata[last_size].coords[nidx*2];
+                        int ny = ssdata[last_size].coords[nidx*2+1];
+
+                        // extract subset
+                        util::extract_ss(ss_ref, nx, ny, px_hori, px_vert, img_ref);
+
+                        // if the neighbouring subset had not met correlation threshold then try values from fft windowing
+                        if (util::cost_arr[idx_results] < opt.opt_threshold){
+                            std::fill(opt.p.begin(), opt.p.end(), 0.0);
+                            opt.p[0] = fourier::shifts[last_size].x[nidx];
+                            opt.p[1] = fourier::shifts[last_size].y[nidx];
+                        }
+                        else {
+                            for (int i = 0; i < opt.num_params; i++){
+                                opt.p[i] = util::p_arr[idx_results_p+i];
+                            }
+                        }
+
+                        // optimize
+                        double centre_x = nx + static_cast<double>(ss_size)/2.0 - 0.5;
+                        double centre_y = ny + static_cast<double>(ss_size)/2.0 - 0.5;
+                        util::Results nres = optimizer::solve(centre_x, centre_y, ss_ref, ss_def, interp_def, opt, conf.corr_crit);
+
+
+                        // if its not SSD, then we need to flip the cost values so that 1.0
+                        // is a perfect match rather than 0.0
+                        if (conf.corr_crit!="SSD")
+                            nres.cost = 1.0-nres.cost;
+
+                        // append results
+                        #pragma omp critical
+                            util::append_results(img_num, nidx, nres, num_ss);
+
+                        // add results to temp neighbour results
+                        temp_neigh.emplace_back(nidx, nres.cost);
+
+                        // update progress bar
+                        int progress = current_progress.fetch_add(1);
+                        util::update_progress_bar(bar, progress, num_ss, prev_pct);
+
+                    }
+                }
+
+                for (const auto& neigh : temp_neigh) {
+                    std::lock_guard<std::mutex> lock(queue_mutexes[tid]);
+                    thread_q.push(neigh);
+                }
+            }
+        }
+        bar.mark_as_completed();
+        indicators::show_console_cursor(true);
+
+    }
+
+
+    void multi_window_fourier(const double *img_ref,
+                              const double *img_def,
+                              const Interpolator &interp_def,
+                              const std::vector<util::SubsetData> &ssdata,
+                              const util::Config &conf,
+                              const int img_num){
+
+        // for the first image perform the FFT windowing. later images will be
+        // seeded with previous images
+        fourier::mgwd(ssdata, img_ref, img_def, interp_def, 
+                      conf.fft_mad, conf.fft_mad_scale);
+
+        const int nsizes = ssdata.size();
+        const int last_size = nsizes-1;
+
+        // get number of subsets and the size for the smalllest window size
+        const int num_ss  = ssdata[last_size].num;
+        const int ss_size = ssdata[last_size].size;
+
+        // progress bar
+        indicators::ProgressBar bar;
+        util::create_progress_bar(bar, conf.filenames, img_num, num_ss);
+        std::atomic<int> current_progress = 0;
+        std::atomic<int> prev_pct = 0;
+
+        // loop over subsets within the ROI
+        #pragma omp parallel shared(stop_request)
+        {
+
+            // initialise subsets
+            util::Subset ss_def(ss_size);
+            util::Subset ss_ref(ss_size);
+
+            // optimization parameters
+            optimizer::Parameters opt(conf.num_params, conf.max_iter, 
+                                    conf.precision, conf.opt_threshold,
+                                    conf.px_vert, conf.px_hori);
+            
+
+            #pragma omp for
+            for (int ss = 0; ss < num_ss; ss++){
+
+                // exit the main DIC loop when ctrl+C is hit
+                if (stop_request){
+                    continue;
+                }
+
+                // subset coordinate list takes central locations. 
+                // Converting to top left corner for optimization routine
+                int ss_x = ssdata[last_size].coords[ss*2];
+                int ss_y = ssdata[last_size].coords[ss*2+1];
+
+                // get the reference subset
+                util::extract_ss(ss_ref, ss_x, ss_y, conf.px_hori, conf.px_vert, img_ref);
+
+                std::fill(opt.p.begin(), opt.p.end(), 0.0);
+                opt.p[0] = fourier::shifts[last_size].x[ss];
+                opt.p[1] = fourier::shifts[last_size].y[ss];
+
+                // perform optimization on subset from deformed image
+                double centre_x = ss_x + static_cast<double>(ss_size)/2.0 - 0.5;
+                double centre_y = ss_y + static_cast<double>(ss_size)/2.0 - 0.5;
+                util::Results res = optimizer::solve(centre_x, centre_y, ss_ref, ss_def, interp_def, opt, conf.corr_crit);
+                
+
+                // if its not SSD, then we need to flip the cost values so that 1.0
+                // is a perfect match rather than 0.0
+                if (conf.corr_crit!="SSD")
+                    res.cost = 1.0-res.cost;
+
+                // append optimization results to results vectors
+                #pragma omp critical
+                    util::append_results(img_num, ss, res, num_ss);
+
+                // update progress bar
+                int progress = current_progress.fetch_add(1);
+                util::update_progress_bar(bar, progress, num_ss, prev_pct);
+
+            }
+        }
+        bar.mark_as_completed();
+        indicators::show_console_cursor(true);
+    }
+
+
+    void single_window_fourier(const double *img_ref,
+                              const double *img_def,
+                              const Interpolator &interp_def,
+                              const util::SubsetData &ssdata,
+                              const util::Config &conf,
+                              const int img_num){
+
+        // for the first image perform the FFT windowing. later images will be
+        // seeded with previous images
+        fourier::sgwd(ssdata, 256, img_ref, img_def, interp_def);
+
+        // get number of subsets and the size for the smalllest window size
+        const int num_ss  = ssdata.num;
+        const int ss_size = ssdata.size;
+
+        // progress bar
+        indicators::ProgressBar bar;
+        util::create_progress_bar(bar, conf.filenames, img_num, num_ss);
+        std::atomic<int> current_progress = 0;
+        std::atomic<int> prev_pct = 0;
+
+        // loop over subsets within the ROI
+        #pragma omp parallel shared(stop_request)
+        {
+
+            // initialise subsets
+            util::Subset ss_def(ss_size);
+            util::Subset ss_ref(ss_size);
+
+            // optimization parameters
+            optimizer::Parameters opt(conf.num_params, conf.max_iter, 
+                                    conf.precision, conf.opt_threshold,
+                                    conf.px_vert, conf.px_hori);
+
+
+            #pragma omp for
+            for (int ss = 0; ss < num_ss; ss++){
+
+                // exit the main DIC loop when ctrl+C is hit
+                if (stop_request){
+                    continue;
+                }
+
+                // subset coordinate list takes central locations. 
+                // Converting to top left corner for optimization routine
+                int ss_x = ssdata.coords[ss*2];
+                int ss_y = ssdata.coords[ss*2+1];
+
+                // get the reference subset
+                util::extract_ss(ss_ref, ss_x, ss_y, conf.px_hori, conf.px_vert, img_ref);
+
+                std::fill(opt.p.begin(), opt.p.end(), 0.0);
+                opt.p[0] = fourier::shifts[0].x[ss];
+                opt.p[1] = fourier::shifts[0].y[ss];
+
+                // perform optimization on subset from deformed image
+                double centre_x = ss_x + static_cast<double>(ss_size)/2.0 - 0.5;
+                double centre_y = ss_y + static_cast<double>(ss_size)/2.0 - 0.5;
+                util::Results res = optimizer::solve(centre_x, centre_y, ss_ref, ss_def, interp_def, opt, conf.corr_crit);
+                
+                // if its not SSD, then we need to flip the cost values so that 1.0
+                // is a perfect match rather than 0.0
+                if (conf.corr_crit!="SSD")
+                    res.cost = 1.0-res.cost;
+
+                // append optimization results to results vectors
+                util::append_results(img_num, ss, res, num_ss);
+
+                // update progress bar
+                int progress = current_progress.fetch_add(1);
+                util::update_progress_bar(bar, progress, num_ss, prev_pct);
+
+            }
+        }
+        bar.mark_as_completed();
+        indicators::show_console_cursor(true);
+    }
+
+}