Add YOLOv5-cls Model (#335)

* add yolov5cls

* fixed bugs

* fixed bugs

* fixed preprocess bug

* add yolov5cls readme

* deal with comments

* Add YOLOv5Cls Note

* add yolov5cls test

Co-authored-by: Jason <jiangjiajun@baidu.com>

fastdeploy/vision.h

@@ -15,6 +15,7 @@
 
 #include "fastdeploy/core/config.h"
 #ifdef ENABLE_VISION
+#include "fastdeploy/vision/classification/contrib/yolov5cls.h"
 #include "fastdeploy/vision/classification/ppcls/model.h"
 #include "fastdeploy/vision/detection/contrib/nanodet_plus.h"
 #include "fastdeploy/vision/detection/contrib/scaledyolov4.h"

fastdeploy/vision/classification/classification_pybind.cc

@@ -16,11 +16,13 @@
 
 namespace fastdeploy {
 
+void BindYOLOv5Cls(pybind11::module& m);
 void BindPaddleClas(pybind11::module& m);
 
 void BindClassification(pybind11::module& m) {
   auto classification_module =
       m.def_submodule("classification", "Image classification models.");
+  BindYOLOv5Cls(classification_module);
   BindPaddleClas(classification_module);
 }
 }  // namespace fastdeploy

fastdeploy/vision/classification/contrib/yolov5cls.cc

@@ -0,0 +1,116 @@
+// 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/vision/classification/contrib/yolov5cls.h"
+
+#include "fastdeploy/utils/perf.h"
+#include "fastdeploy/vision/utils/utils.h"
+
+namespace fastdeploy {
+namespace vision {
+namespace classification {
+
+YOLOv5Cls::YOLOv5Cls(const std::string& model_file,
+                     const std::string& params_file,
+                     const RuntimeOption& custom_option,
+                     const ModelFormat& model_format) {
+  if (model_format == ModelFormat::ONNX) {
+    valid_cpu_backends = {Backend::OPENVINO, Backend::ORT};
+    valid_gpu_backends = {Backend::ORT, Backend::TRT};
+  } else {
+    valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
+    valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+  }
+  runtime_option = custom_option;
+  runtime_option.model_format = model_format;
+  runtime_option.model_file = model_file;
+  runtime_option.params_file = params_file;
+  initialized = Initialize();
+}
+
+bool YOLOv5Cls::Initialize() {
+  // preprocess parameters
+  size = {224, 224};
+  if (!InitRuntime()) {
+    FDERROR << "Failed to initialize fastdeploy backend." << std::endl;
+    return false;
+  }
+  return true;
+}
+
+bool YOLOv5Cls::Preprocess(Mat* mat, FDTensor* output,
+                           const std::vector<int>& size) {
+  // CenterCrop
+  int crop_size = std::min(mat->Height(), mat->Width());
+  CenterCrop::Run(mat, crop_size, crop_size);
+  Resize::Run(mat, size[0], size[1], -1, -1, cv::INTER_LINEAR);
+  // Normalize
+  BGR2RGB::Run(mat);
+  std::vector<float> alpha = {1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f};
+  std::vector<float> beta = {0.0f, 0.0f, 0.0f};
+  Convert::Run(mat, alpha, beta);
+  std::vector<float> mean = {0.485f, 0.456f, 0.406f};
+  std::vector<float> std = {0.229f, 0.224f, 0.225f};
+  Normalize::Run(mat, mean, std, false);
+  HWC2CHW::Run(mat);
+  Cast::Run(mat, "float");
+
+  mat->ShareWithTensor(output);
+  output->shape.insert(output->shape.begin(), 1);
+  return true;
+}
+
+bool YOLOv5Cls::Postprocess(const FDTensor& infer_result,
+                            ClassifyResult* result, int topk) {
+  // Softmax
+  FDTensor infer_result_softmax;
+  Softmax(infer_result, &infer_result_softmax, 1);
+  int num_classes = infer_result_softmax.shape[1];
+  const float* infer_result_buffer =
+      reinterpret_cast<const float*>(infer_result_softmax.Data());
+  topk = std::min(num_classes, topk);
+  result->label_ids =
+      utils::TopKIndices(infer_result_buffer, num_classes, topk);
+  result->scores.resize(topk);
+  for (int i = 0; i < topk; ++i) {
+    result->scores[i] = *(infer_result_buffer + result->label_ids[i]);
+  }
+  return true;
+}
+
+bool YOLOv5Cls::Predict(cv::Mat* im, ClassifyResult* result, int topk) {
+  Mat mat(*im);
+  std::vector<FDTensor> input_tensors(1);
+  if (!Preprocess(&mat, &input_tensors[0], size)) {
+    FDERROR << "Failed to preprocess input image." << std::endl;
+    return false;
+  }
+
+  input_tensors[0].name = InputInfoOfRuntime(0).name;
+  std::vector<FDTensor> output_tensors(1);
+  if (!Infer(input_tensors, &output_tensors)) {
+    FDERROR << "Failed to inference." << std::endl;
+    return false;
+  }
+
+  if (!Postprocess(output_tensors[0], result, topk)) {
+    FDERROR << "Failed to post process." << std::endl;
+    return false;
+  }
+  return true;
+}
+
+}  // namespace classification
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/classification/contrib/yolov5cls.h

@@ -0,0 +1,70 @@
+// 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.
+
+#pragma once
+#include "fastdeploy/fastdeploy_model.h"
+#include "fastdeploy/vision/common/processors/transform.h"
+#include "fastdeploy/vision/common/result.h"
+
+namespace fastdeploy {
+namespace vision {
+/** \brief All image classification model APIs are defined inside this namespace
+ *
+ */
+namespace classification {
+
+/*! @brief YOLOv5Cls model object used when to load a YOLOv5Cls model exported by YOLOv5
+ */
+class FASTDEPLOY_DECL YOLOv5Cls : public FastDeployModel {
+ public:
+  /** \brief Set path of model file and configuration file, and the configuration of runtime
+   *
+   * \param[in] model_file Path of model file, e.g yolov5cls/yolov5n-cls.onnx
+   * \param[in] params_file Path of parameter file, if the model format is ONNX, this parameter will be ignored
+   * \param[in] custom_option RuntimeOption for inference, the default will use cpu, and choose the backend defined in `valid_cpu_backends`
+   * \param[in] model_format Model format of the loaded model, default is ONNX format
+   */
+  YOLOv5Cls(const std::string& model_file, const std::string& params_file = "",
+            const RuntimeOption& custom_option = RuntimeOption(),
+            const ModelFormat& model_format = ModelFormat::ONNX);
+
+  /// Get model's name
+  virtual std::string ModelName() const { return "yolov5cls"; }
+
+  /** \brief Predict the classification result for an input image
+   *
+   * \param[in] im The input image data, comes from cv::imread()
+   * \param[in] result The output classification result will be writen to this structure
+   * \param[in] topk Returns the topk classification result with the highest predicted probability, the default is 1
+   * \return true if the prediction successed, otherwise false
+   */
+  virtual bool Predict(cv::Mat* im, ClassifyResult* result, int topk = 1);
+
+  /// Preprocess image size, the default is (224, 224)
+  std::vector<int> size;
+
+ private:
+  bool Initialize();
+  /// Preprocess an input image, and set the preprocessed results to `outputs`
+  bool Preprocess(Mat* mat, FDTensor* output,
+                  const std::vector<int>& size = {224, 224});
+
+  /// Postprocess the inferenced results, and set the final result to `result`
+  bool Postprocess(const FDTensor& infer_result, ClassifyResult* result,
+                   int topk = 1);
+};
+
+}  // namespace classification
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/classification/contrib/yolov5cls_pybind.cc

@@ -0,0 +1,32 @@
+// 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 BindYOLOv5Cls(pybind11::module& m) {
+  pybind11::class_<vision::classification::YOLOv5Cls, FastDeployModel>(
+      m, "YOLOv5Cls")
+      .def(pybind11::init<std::string, std::string, RuntimeOption,
+                          ModelFormat>())
+      .def("predict",
+           [](vision::classification::YOLOv5Cls& self, pybind11::array& data,
+              int topk = 1) {
+             auto mat = PyArrayToCvMat(data);
+             vision::ClassifyResult res;
+             self.Predict(&mat, &res, topk);
+             return res;
+           })
+      .def_readwrite("size", &vision::classification::YOLOv5Cls::size);
+}
+}  // namespace fastdeploy