Deploee

csrc/mmdeploy/operation/cuda/cvtcolor.cpp

@@ -90,19 +90,20 @@ class CvtColorImpl : public CvtColor {
 
     auto height = src.height();
     auto width = src.width();
-    auto channels = src.channel();
-    auto stride = width * channels;
+    auto src_channels = src.channel();
+    auto src_stride = width * src_channels;
 
     Mat dst_mat(height, width, dst_fmt, src.type(), device());
+    auto dst_stride = width * dst_mat.channel();
 
     auto convert = [&](auto type) -> Result<void> {
       using T = typename decltype(type)::type;
       auto converter = GetConverter<T>(src.pixel_format(), dst_fmt);
       if (!converter) {
         return Status(eNotSupported);
       }
-      auto ret =
-          converter(cuda_stream, height, width, stride, src.data<T>(), stride, dst_mat.data<T>());
+      auto ret = converter(cuda_stream, height, width, src_stride, src.data<T>(), dst_stride,
+                           dst_mat.data<T>());
       if (ret != ppl::common::RC_SUCCESS) {
         return Status(eFail);
       }

csrc/mmdeploy/preprocess/transform/load.cpp

@@ -55,7 +55,7 @@ class PrepareImage : public Transform {
     if (color_type_ == "color" || color_type_ == "color_ignore_orientation") {
       OUTCOME_TRY(cvt_color_.Apply(src_mat, dst_mat, PixelFormat::kBGR));
     } else {
-      OUTCOME_TRY(cvt_color_.Apply(dst_mat, dst_mat, PixelFormat::kGRAYSCALE));
+      OUTCOME_TRY(cvt_color_.Apply(src_mat, dst_mat, PixelFormat::kGRAYSCALE));
     }
     auto tensor = to_tensor(dst_mat);
     if (to_float32_) {

demo/csrc/CMakeLists.txt

@@ -41,6 +41,8 @@ if (MMDEPLOY_BUILD_SDK_CXX_API)
     add_example(restorer cpp restorer)
     add_example(text_detector cpp text_ocr)
     add_example(text_detector cpp text_det_recog)
+    add_example(text_detector cpp text_det)
+    add_example(text_recognizer cpp text_recog)
     add_example(pose_detector cpp pose_detector)
     add_example(rotated_detector cpp rotated_detector)
     add_example(pose_detector cpp pose_tracker)

demo/csrc/cpp/text_det.cxx

@@ -0,0 +1,47 @@
+
+#include <opencv2/imgcodecs/imgcodecs.hpp>
+#include <opencv2/imgproc/imgproc.hpp>
+#include <string>
+
+#include "mmdeploy/common.hpp"
+#include "mmdeploy/text_detector.hpp"
+
+int main(int argc, char* argv[]) {
+  if (argc != 4) {
+    fprintf(stderr, "usage:\n  text_det device_name model_path image_path\n");
+    return 1;
+  }
+  auto device_name = argv[1];
+  auto model_path = argv[2];
+  auto image_path = argv[3];
+  cv::Mat img = cv::imread(image_path);
+  if (!img.data) {
+    fprintf(stderr, "failed to load image: %s\n", image_path);
+    return 1;
+  }
+
+  mmdeploy::Profiler profiler{"/tmp/profile.bin"};
+  mmdeploy::Context context(mmdeploy::Device(device_name, 0));
+  context.Add(profiler);
+
+  mmdeploy::Model model(model_path);
+  mmdeploy::TextDetector detector(model, context);
+
+  const int REPEAT = 20;
+  auto res = detector.Apply(img);
+  for (int i = 0; i < REPEAT - 1; ++i) {
+    res = detector.Apply(img);
+  }
+
+  for (const auto& det : res) {
+    const auto& box = det.bbox;
+    for (int i = 0; i < 4; i++) {
+      cv::rectangle(img, cv::Point{(int)box[i].x, (int)box[i].y},
+                    cv::Point{(int)box[(i + 1) % 4].x, (int)box[(i + 1) % 4].y},
+                    cv::Scalar{0, 255, 0});
+    }
+  }
+  cv::imwrite("output_ocr_detection.jpg", img);
+
+  return 0;
+}

demo/csrc/cpp/text_recog.cxx

@@ -0,0 +1,52 @@
+
+#include <opencv2/imgcodecs/imgcodecs.hpp>
+#include <opencv2/imgproc/imgproc.hpp>
+#include <string>
+#include <vector>
+
+#include "mmdeploy/common.hpp"
+#include "mmdeploy/text_detector.hpp"
+#include "mmdeploy/text_recognizer.hpp"
+
+int main(int argc, char* argv[]) {
+  if (argc != 4) {
+    fprintf(stderr, "usage:\n  text_recog device_name model_path image_path\n");
+    return 1;
+  }
+  auto device_name = argv[1];
+  auto model_path = argv[2];
+  auto image_path = argv[3];
+  cv::Mat img = cv::imread(image_path);
+  if (!img.data) {
+    fprintf(stderr, "failed to load image: %s\n", image_path);
+    return 1;
+  }
+
+  mmdeploy::TextDetection bbox = {{{0.f, (float)img.rows - 1},
+                                   {0.f, 0.f},
+                                   {(float)img.cols - 1, 0},
+                                   {(float)img.cols, (float)img.rows - 1}},
+                                  1.0f};
+  std::vector<mmdeploy::TextDetection> bboxes = {bbox};
+
+  mmdeploy::Profiler profiler{"/tmp/profile.bin"};
+  mmdeploy::Context context(mmdeploy::Device(device_name, 0));
+  context.Add(profiler);
+
+  mmdeploy::Model model(model_path);
+  mmdeploy::TextRecognizer recognizer(model, context);
+
+  const int REPEAT = 20;
+  auto res = recognizer.Apply(img, bboxes);
+  for (int i = 0; i < REPEAT - 1; ++i) {
+    res = recognizer.Apply(img);
+  }
+
+  for (auto& reg : res) {
+    for (int i = 0; i < reg.length; i++) {
+      printf("%c %f\n", reg.text[i], reg.score[i]);
+    }
+  }
+
+  return 0;
+}