PaddlePaddle · jiangjiajun · Aug 9, 2022 · Aug 8, 2022 · Aug 9, 2022
diff --git a/csrcs/fastdeploy/pybind/main.cc.in b/csrcs/fastdeploy/pybind/main.cc.in
@@ -30,6 +30,8 @@ pybind11::dtype FDDataTypeToNumpyDataType(const FDDataType& fd_dtype) {
     dt = pybind11::dtype::of<float>();
   } else if (fd_dtype == FDDataType::FP64) {
     dt = pybind11::dtype::of<double>();
+  } else if (fd_dtype == FDDataType::UINT8) {
+    dt = pybind11::dtype::of<uint8_t>();
   } else {
     FDASSERT(false, "The function doesn't support data type of " +
                         Str(fd_dtype) + ".");
@@ -46,6 +48,8 @@ FDDataType NumpyDataTypeToFDDataType(const pybind11::dtype& np_dtype) {
     return FDDataType::FP32;
   } else if (np_dtype.is(pybind11::dtype::of<double>())) {
     return FDDataType::FP64;
+  } else if (np_dtype.is(pybind11::dtype::of<uint8_t>())) {
+    return FDDataType::UINT8;
   }
   FDASSERT(false,
            "NumpyDataTypeToFDDataType() only support "
@@ -66,6 +70,13 @@ void PyArrayToTensor(pybind11::array& pyarray, FDTensor* tensor,
   }
 }
 
+pybind11::array TensorToPyArray(const FDTensor& tensor) {
+  auto numpy_dtype = FDDataTypeToNumpyDataType(tensor.dtype);
+  auto out = pybind11::array(numpy_dtype, tensor.shape);
+  memcpy(out.mutable_data(), tensor.Data(), tensor.Numel() * FDDataTypeSize(tensor.dtype));
+  return out;
+}
+
 #ifdef ENABLE_VISION
 int NumpyDataTypeToOpenCvType(const pybind11::dtype& np_dtype) {
   if (np_dtype.is(pybind11::dtype::of<int32_t>())) {

diff --git a/csrcs/fastdeploy/pybind/main.h b/csrcs/fastdeploy/pybind/main.h
@@ -36,12 +36,14 @@ FDDataType NumpyDataTypeToFDDataType(const pybind11::dtype& np_dtype);
 
 void PyArrayToTensor(pybind11::array& pyarray, FDTensor* tensor,
                      bool share_buffer = false);
+pybind11::array TensorToPyArray(const FDTensor& tensor);
 
 #ifdef ENABLE_VISION
 cv::Mat PyArrayToCvMat(pybind11::array& pyarray);
 #endif
 
-template <typename T> FDDataType CTypeToFDDataType() {
+template <typename T>
+FDDataType CTypeToFDDataType() {
   if (std::is_same<T, int32_t>::value) {
     return FDDataType::INT32;
   } else if (std::is_same<T, int64_t>::value) {
@@ -57,9 +59,9 @@ template <typename T> FDDataType CTypeToFDDataType() {
 }
 
 template <typename T>
-std::vector<pybind11::array>
-PyBackendInfer(T& self, const std::vector<std::string>& names,
-               std::vector<pybind11::array>& data) {
+std::vector<pybind11::array> PyBackendInfer(
+    T& self, const std::vector<std::string>& names,
+    std::vector<pybind11::array>& data) {
   std::vector<FDTensor> inputs(data.size());
   for (size_t i = 0; i < data.size(); ++i) {
     // TODO(jiangjiajun) here is considered to use user memory directly
@@ -85,4 +87,4 @@ PyBackendInfer(T& self, const std::vector<std::string>& names,
   return results;
 }
 
-} // namespace fastdeploy
+}  // namespace fastdeploy
diff --git a/csrcs/fastdeploy/vision.h b/csrcs/fastdeploy/vision.h
@@ -23,6 +23,9 @@
 #include "fastdeploy/vision/deepinsight/partial_fc.h"
 #include "fastdeploy/vision/deepinsight/scrfd.h"
 #include "fastdeploy/vision/deepinsight/vpl.h"
+#include "fastdeploy/vision/detection/contrib/scaledyolov4.h"
+#include "fastdeploy/vision/detection/contrib/yolor.h"
+#include "fastdeploy/vision/detection/contrib/yolov7.h"
 #include "fastdeploy/vision/linzaer/ultraface.h"
 #include "fastdeploy/vision/megvii/yolox.h"
 #include "fastdeploy/vision/meituan/yolov6.h"
@@ -32,9 +35,6 @@
 #include "fastdeploy/vision/ppseg/model.h"
 #include "fastdeploy/vision/rangilyu/nanodet_plus.h"
 #include "fastdeploy/vision/ultralytics/yolov5.h"
-#include "fastdeploy/vision/wongkinyiu/scaledyolov4.h"
-#include "fastdeploy/vision/wongkinyiu/yolor.h"
-#include "fastdeploy/vision/wongkinyiu/yolov7.h"
 #include "fastdeploy/vision/zhkkke/modnet.h"
 #endif
 

diff --git a/csrcs/fastdeploy/vision/common/processors/mat.h b/csrcs/fastdeploy/vision/common/processors/mat.h
@@ -27,7 +27,7 @@ namespace vision {
 
 enum Layout { HWC, CHW };
 
-struct Mat {
+struct FASTDEPLOY_DECL Mat {
   explicit Mat(cv::Mat& mat) {
     cpu_mat = mat;
     device = Device::CPU;
@@ -76,5 +76,5 @@ struct Mat {
   Device device = Device::CPU;
 };
 
-} // namespace vision
-} // namespace fastdeploy
+}  // namespace vision
+}  // namespace fastdeploy
diff --git a/...tdeploy/vision/wongkinyiu/scaledyolov4.cc → .../vision/detection/contrib/scaledyolov4.cc b/...tdeploy/vision/wongkinyiu/scaledyolov4.cc → .../vision/detection/contrib/scaledyolov4.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "fastdeploy/vision/wongkinyiu/scaledyolov4.h"
+#include "fastdeploy/vision/detection/contrib/scaledyolov4.h"
 #include "fastdeploy/utils/perf.h"
 #include "fastdeploy/vision/utils/utils.h"
 
 namespace fastdeploy {
 namespace vision {
-namespace wongkinyiu {
+namespace detection {
 
 void ScaledYOLOv4::LetterBox(Mat* mat, const std::vector<int>& size,
                              const std::vector<float>& color, bool _auto,
@@ -65,8 +65,8 @@ ScaledYOLOv4::ScaledYOLOv4(const std::string& model_file,
     valid_cpu_backends = {Backend::ORT};  // 指定可用的CPU后端
     valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
   } else {
-    valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
-    valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_cpu_backends = {Backend::PDINFER};
+    valid_gpu_backends = {Backend::PDINFER};
   }
   runtime_option = custom_option;
   runtime_option.model_format = model_format;
@@ -219,10 +219,6 @@ bool ScaledYOLOv4::Postprocess(
 
 bool ScaledYOLOv4::Predict(cv::Mat* im, DetectionResult* result,
                            float conf_threshold, float nms_iou_threshold) {
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_START(0)
-#endif
-
   Mat mat(*im);
   std::vector<FDTensor> input_tensors(1);
 
@@ -239,34 +235,21 @@ bool ScaledYOLOv4::Predict(cv::Mat* im, DetectionResult* result,
     return false;
   }
 
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_END(0, "Preprocess")
-  TIMERECORD_START(1)
-#endif
-
   input_tensors[0].name = InputInfoOfRuntime(0).name;
   std::vector<FDTensor> output_tensors;
   if (!Infer(input_tensors, &output_tensors)) {
     FDERROR << "Failed to inference." << std::endl;
     return false;
   }
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_END(1, "Inference")
-  TIMERECORD_START(2)
-#endif
-
   if (!Postprocess(output_tensors[0], result, im_info, conf_threshold,
                    nms_iou_threshold)) {
     FDERROR << "Failed to post process." << std::endl;
     return false;
   }
 
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_END(2, "Postprocess")
-#endif
   return true;
 }
 
-}  // namespace wongkinyiu
+}  // namespace detection
 }  // namespace vision
 }  // namespace fastdeploy
diff --git a/...stdeploy/vision/wongkinyiu/scaledyolov4.h → ...y/vision/detection/contrib/scaledyolov4.h b/...stdeploy/vision/wongkinyiu/scaledyolov4.h → ...y/vision/detection/contrib/scaledyolov4.h
@@ -19,7 +19,7 @@
 
 namespace fastdeploy {
 namespace vision {
-namespace wongkinyiu {
+namespace detection {
 
 class FASTDEPLOY_DECL ScaledYOLOv4 : public FastDeployModel {
  public:
@@ -31,7 +31,7 @@ class FASTDEPLOY_DECL ScaledYOLOv4 : public FastDeployModel {
                const Frontend& model_format = Frontend::ONNX);
 
   // 定义模型的名称
-  virtual std::string ModelName() const { return "WongKinYiu/ScaledYOLOv4"; }
+  virtual std::string ModelName() const { return "ScaledYOLOv4"; }
 
   // 模型预测接口，即用户调用的接口
   // im 为用户的输入数据，目前对于CV均定义为cv::Mat
@@ -98,6 +98,6 @@ class FASTDEPLOY_DECL ScaledYOLOv4 : public FastDeployModel {
   // auto check by fastdeploy after the internal Runtime already initialized.
   bool is_dynamic_input_;
 };
-}  // namespace wongkinyiu
+}  // namespace detection
 }  // namespace vision
 }  // namespace fastdeploy
diff --git a/csrcs/fastdeploy/vision/detection/contrib/scaledyolov4_pybind.cc b/csrcs/fastdeploy/vision/detection/contrib/scaledyolov4_pybind.cc
@@ -0,0 +1,41 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/pybind/main.h"
+
+namespace fastdeploy {
+void BindScaledYOLOv4(pybind11::module& m) {
+  pybind11::class_<vision::detection::ScaledYOLOv4, FastDeployModel>(
+      m, "ScaledYOLOv4")
+      .def(pybind11::init<std::string, std::string, RuntimeOption, Frontend>())
+      .def("predict",
+           [](vision::detection::ScaledYOLOv4& self, pybind11::array& data,
+              float conf_threshold, float nms_iou_threshold) {
+             auto mat = PyArrayToCvMat(data);
+             vision::DetectionResult res;
+             self.Predict(&mat, &res, conf_threshold, nms_iou_threshold);
+             return res;
+           })
+      .def_readwrite("size", &vision::detection::ScaledYOLOv4::size)
+      .def_readwrite("padding_value",
+                     &vision::detection::ScaledYOLOv4::padding_value)
+      .def_readwrite("is_mini_pad",
+                     &vision::detection::ScaledYOLOv4::is_mini_pad)
+      .def_readwrite("is_no_pad", &vision::detection::ScaledYOLOv4::is_no_pad)
+      .def_readwrite("is_scale_up",
+                     &vision::detection::ScaledYOLOv4::is_scale_up)
+      .def_readwrite("stride", &vision::detection::ScaledYOLOv4::stride)
+      .def_readwrite("max_wh", &vision::detection::ScaledYOLOv4::max_wh);
+}
+}  // namespace fastdeploy
diff --git a/csrcs/fastdeploy/vision/wongkinyiu/yolor.cc → ...tdeploy/vision/detection/contrib/yolor.cc b/csrcs/fastdeploy/vision/wongkinyiu/yolor.cc → ...tdeploy/vision/detection/contrib/yolor.cc
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "fastdeploy/vision/wongkinyiu/yolor.h"
+#include "fastdeploy/vision/detection/contrib/yolor.h"
 #include "fastdeploy/utils/perf.h"
 #include "fastdeploy/vision/utils/utils.h"
 
 namespace fastdeploy {
 namespace vision {
-namespace wongkinyiu {
+namespace detection {
 
 void YOLOR::LetterBox(Mat* mat, const std::vector<int>& size,
                       const std::vector<float>& color, bool _auto,
@@ -63,8 +63,8 @@ YOLOR::YOLOR(const std::string& model_file, const std::string& params_file,
     valid_cpu_backends = {Backend::ORT};  // 指定可用的CPU后端
     valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
   } else {
-    valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
-    valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_cpu_backends = {Backend::PDINFER};
+    valid_gpu_backends = {Backend::PDINFER};
   }
   runtime_option = custom_option;
   runtime_option.model_format = model_format;
@@ -216,10 +216,6 @@ bool YOLOR::Postprocess(
 
 bool YOLOR::Predict(cv::Mat* im, DetectionResult* result, float conf_threshold,
                     float nms_iou_threshold) {
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_START(0)
-#endif
-
   Mat mat(*im);
   std::vector<FDTensor> input_tensors(1);
 
@@ -236,34 +232,22 @@ bool YOLOR::Predict(cv::Mat* im, DetectionResult* result, float conf_threshold,
     return false;
   }
 
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_END(0, "Preprocess")
-  TIMERECORD_START(1)
-#endif
-
   input_tensors[0].name = InputInfoOfRuntime(0).name;
   std::vector<FDTensor> output_tensors;
   if (!Infer(input_tensors, &output_tensors)) {
     FDERROR << "Failed to inference." << std::endl;
     return false;
   }
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_END(1, "Inference")
-  TIMERECORD_START(2)
-#endif
 
   if (!Postprocess(output_tensors[0], result, im_info, conf_threshold,
                    nms_iou_threshold)) {
     FDERROR << "Failed to post process." << std::endl;
     return false;
   }
 
-#ifdef FASTDEPLOY_DEBUG
-  TIMERECORD_END(2, "Postprocess")
-#endif
   return true;
 }
 
-}  // namespace wongkinyiu
+}  // namespace detection
 }  // namespace vision
 }  // namespace fastdeploy
diff --git a/csrcs/fastdeploy/vision/wongkinyiu/yolor.h → ...stdeploy/vision/detection/contrib/yolor.h b/csrcs/fastdeploy/vision/wongkinyiu/yolor.h → ...stdeploy/vision/detection/contrib/yolor.h
@@ -19,7 +19,7 @@
 
 namespace fastdeploy {
 namespace vision {
-namespace wongkinyiu {
+namespace detection {
 
 class FASTDEPLOY_DECL YOLOR : public FastDeployModel {
  public:
@@ -30,7 +30,7 @@ class FASTDEPLOY_DECL YOLOR : public FastDeployModel {
         const Frontend& model_format = Frontend::ONNX);
 
   // 定义模型的名称
-  virtual std::string ModelName() const { return "WongKinYiu/yolor"; }
+  virtual std::string ModelName() const { return "YOLOR"; }
 
   // 模型预测接口，即用户调用的接口
   // im 为用户的输入数据，目前对于CV均定义为cv::Mat
@@ -97,6 +97,6 @@ class FASTDEPLOY_DECL YOLOR : public FastDeployModel {
   // auto check by fastdeploy after the internal Runtime already initialized.
   bool is_dynamic_input_;
 };
-}  // namespace wongkinyiu
+}  // namespace detection
 }  // namespace vision
 }  // namespace fastdeploy
diff --git a/csrcs/fastdeploy/vision/detection/contrib/yolor_pybind.cc b/csrcs/fastdeploy/vision/detection/contrib/yolor_pybind.cc
@@ -0,0 +1,37 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/pybind/main.h"
+
+namespace fastdeploy {
+void BindYOLOR(pybind11::module& m) {
+  pybind11::class_<vision::detection::YOLOR, FastDeployModel>(m, "YOLOR")
+      .def(pybind11::init<std::string, std::string, RuntimeOption, Frontend>())
+      .def("predict",
+           [](vision::detection::YOLOR& self, pybind11::array& data,
+              float conf_threshold, float nms_iou_threshold) {
+             auto mat = PyArrayToCvMat(data);
+             vision::DetectionResult res;
+             self.Predict(&mat, &res, conf_threshold, nms_iou_threshold);
+             return res;
+           })
+      .def_readwrite("size", &vision::detection::YOLOR::size)
+      .def_readwrite("padding_value", &vision::detection::YOLOR::padding_value)
+      .def_readwrite("is_mini_pad", &vision::detection::YOLOR::is_mini_pad)
+      .def_readwrite("is_no_pad", &vision::detection::YOLOR::is_no_pad)
+      .def_readwrite("is_scale_up", &vision::detection::YOLOR::is_scale_up)
+      .def_readwrite("stride", &vision::detection::YOLOR::stride)
+      .def_readwrite("max_wh", &vision::detection::YOLOR::max_wh);
+}
+}  // namespace fastdeploy