Improve ALPR_OCR peformance

modules/ANSOCR/ANSONNXOCR/ONNXOCRDetector.cpp

@@ -7,6 +7,7 @@
 #include <iostream>
 #include <algorithm>
 #include <cmath>
+#include <chrono>
 
 namespace ANSCENTER {
 namespace onnxocr {
@@ -15,6 +16,12 @@ ONNXOCRDetector::ONNXOCRDetector(const std::string& onnx_path, unsigned int num_
     : BasicOrtHandler(onnx_path, num_threads) {
 }
 
+ONNXOCRDetector::ONNXOCRDetector(const std::string& onnx_path,
+                                 const OrtHandlerOptions& options,
+                                 unsigned int num_threads)
+    : BasicOrtHandler(onnx_path, options, num_threads) {
+}
+
 Ort::Value ONNXOCRDetector::transform(const cv::Mat& mat) {
     // Not used directly - detection uses custom Preprocess + manual tensor creation
     // Provided to satisfy BasicOrtHandler pure virtual
@@ -308,5 +315,41 @@ std::vector<cv::Point2f> ONNXOCRDetector::UnclipPolygon(const std::array<cv::Poi
     return result;
 }
 
+void ONNXOCRDetector::Warmup() {
+    std::lock_guard<std::mutex> lock(_mutex);
+    if (_warmedUp || !ort_session) return;
+
+    // 320x320 covers the typical license-plate ROI after LPD crop +
+    // multiple-of-32 rounding. cuDNN caches the algorithm for this
+    // shape so the first real inference doesn't pay the picker cost.
+    constexpr int kWarmupSide = 320;
+    try {
+        cv::Mat dummy(kWarmupSide, kWarmupSide, CV_8UC3, cv::Scalar(128, 128, 128));
+        cv::Mat dummyF;
+        dummy.convertTo(dummyF, CV_32FC3);
+        auto inputData = NormalizeAndPermute(dummyF);
+
+        std::array<int64_t, 4> inputShape = { 1, 3, kWarmupSide, kWarmupSide };
+        Ort::Value inputTensor = Ort::Value::CreateTensor<float>(
+            *memory_info_handler, inputData.data(), inputData.size(),
+            inputShape.data(), inputShape.size());
+
+        auto t0 = std::chrono::high_resolution_clock::now();
+        (void)ort_session->Run(
+            Ort::RunOptions{ nullptr },
+            input_node_names.data(), &inputTensor, 1,
+            output_node_names.data(), num_outputs);
+        auto t1 = std::chrono::high_resolution_clock::now();
+        double ms = std::chrono::duration<double, std::milli>(t1 - t0).count();
+        std::cout << "[ONNXOCRDetector] Warmup [1,3,"
+                  << kWarmupSide << "," << kWarmupSide << "] "
+                  << ms << " ms" << std::endl;
+    }
+    catch (const Ort::Exception& e) {
+        std::cerr << "[ONNXOCRDetector] Warmup failed: " << e.what() << std::endl;
+    }
+    _warmedUp = true;
+}
+
 } // namespace onnxocr
 } // namespace ANSCENTER