fixed Ubuntu+gcc build error for std::xxxf (#81)

lite/mnn/cv/mnn_nanodet.cpp

@@ -44,7 +44,7 @@ void MNNNanoDet::resize_unscale(const cv::Mat &mat, cv::Mat &mat_rs,
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -213,10 +213,10 @@ void MNNNanoDet::generate_bboxes_single_stride(const NanoScaleParams &scale_para
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/mnn/cv/mnn_nanodet_efficientnet_lite.cpp

@@ -45,7 +45,7 @@ void MNNNanoDetEfficientNetLite::resize_unscale(const cv::Mat &mat, cv::Mat &mat
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -214,10 +214,10 @@ void MNNNanoDetEfficientNetLite::generate_bboxes_single_stride(const NanoLiteSca
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/mnn/cv/mnn_rvm.h

@@ -14,7 +14,7 @@ namespace mnncv
   public:
     explicit MNNRobustVideoMatting(const std::string &_mnn_path,
                                    unsigned int _num_threads = 1,
-                                   unsigned int _variant_type = VARIANT::MOBILENETV3); //
+                                   unsigned int _variant_type = 0); //
     ~MNNRobustVideoMatting();
 
   private:

lite/mnn/cv/mnn_yolop.cpp

@@ -43,7 +43,7 @@ void MNNYOLOP::resize_unscale(const cv::Mat &mat, cv::Mat &mat_rs,
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor

lite/mnn/cv/mnn_yolox.cpp

@@ -136,8 +136,8 @@ void MNNYoloX::generate_bboxes(std::vector<types::Boxf> &bbox_collection,
 
     float cx = (dx + (float) grid0) * (float) stride;
     float cy = (dy + (float) grid1) * (float) stride;
-    float w = std::expf(dw) * (float) stride;
-    float h = std::expf(dh) * (float) stride;
+    float w = std::exp(dw) * (float) stride;
+    float h = std::exp(dh) * (float) stride;
 
     types::Boxf box;
     box.x1 = (cx - w / 2.f) * scale_width;

lite/ncnn/cv/ncnn_nanodet.cpp

@@ -30,7 +30,7 @@ void NCNNNanoDet::resize_unscale(const cv::Mat &mat, cv::Mat &mat_rs,
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -206,10 +206,10 @@ void NCNNNanoDet::generate_bboxes_single_stride(const NanoScaleParams &scale_par
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/ncnn/cv/ncnn_nanodet_depreciated.cpp

@@ -30,7 +30,7 @@ void NCNNNanoDetDepreciated::resize_unscale(const cv::Mat &mat, cv::Mat &mat_rs,
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -220,10 +220,10 @@ void NCNNNanoDetDepreciated::generate_bboxes_single_stride(const NanoDepreciated
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/ncnn/cv/ncnn_nanodet_efficientnet_lite.cpp

@@ -30,7 +30,7 @@ void NCNNNanoDetEfficientNetLite::resize_unscale(const cv::Mat &mat, cv::Mat &ma
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -206,10 +206,10 @@ void NCNNNanoDetEfficientNetLite::generate_bboxes_single_stride(const NanoLiteSc
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/ncnn/cv/ncnn_nanodet_efficientnet_lite_depreciated.cpp

@@ -32,7 +32,7 @@ void NCNNNanoDetEfficientNetLiteDepreciated::resize_unscale(
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -226,10 +226,10 @@ void NCNNNanoDetEfficientNetLiteDepreciated::generate_bboxes_single_stride(
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/ncnn/cv/ncnn_rvm.h

@@ -17,7 +17,7 @@ namespace ncnncv
                                     unsigned int _num_threads = 1,
                                     int _input_height = 480,
                                     int _input_width = 640,
-                                    unsigned int _variant_type = VARIANT::MOBILENETV3); //
+                                    unsigned int _variant_type = 0); //
     ~NCNNRobustVideoMatting() override = default;
 
   private:

lite/ort/cv/efficientdet.cpp

@@ -143,8 +143,8 @@ void EfficientDet::generate_bboxes(std::vector<types::Boxf> &bbox_collection,
 
     float cx = dx * wa + cxa;
     float cy = dy * ha + cya;
-    float w = std::expf(dw) * wa;
-    float h = std::expf(dh) * ha;
+    float w = std::exp(dw) * wa;
+    float h = std::exp(dh) * ha;
 
     types::Boxf box;
     box.x1 = (cx - w / 2.f) * scale_width;

lite/ort/cv/efficientdet_d7.cpp

@@ -143,8 +143,8 @@ void EfficientDetD7::generate_bboxes(std::vector<types::Boxf> &bbox_collection,
 
     float cx = dx * wa + cxa;
     float cy = dy * ha + cya;
-    float w = std::expf(dw) * wa;
-    float h = std::expf(dh) * ha;
+    float w = std::exp(dw) * wa;
+    float h = std::exp(dh) * ha;
 
     types::Boxf box;
     box.x1 = (cx - w / 2.f) * scale_width;

lite/ort/cv/efficientdet_d8.cpp

@@ -143,8 +143,8 @@ void EfficientDetD8::generate_bboxes(std::vector<types::Boxf> &bbox_collection,
 
     float cx = dx * wa + cxa;
     float cy = dy * ha + cya;
-    float w = std::expf(dw) * wa;
-    float h = std::expf(dh) * ha;
+    float w = std::exp(dw) * wa;
+    float h = std::exp(dh) * ha;
 
     types::Boxf box;
     box.x1 = (cx - w / 2.f) * scale_width;

lite/ort/cv/face_landmarks_1000.cpp

@@ -43,8 +43,8 @@ void FaceLandmark1000::detect(const cv::Mat &mat, types::Landmarks &landmarks)
     float x = landmarks_norm.At<float>({0, i});
     float y = landmarks_norm.At<float>({0, i + 1});
 
-    x = std::fmin(std::fmax(0., x), 1.0f);
-    y = std::fmin(std::fmax(0., y), 1.0f);
+    x = std::min(std::max(0.f, x), 1.0f);
+    y = std::min(std::max(0.f, y), 1.0f);
 
     landmarks.points.push_back(cv::Point2f(x * img_width, y * img_height));
   }

lite/ort/cv/faceboxes.cpp

@@ -52,8 +52,8 @@ void FaceBoxes::generate_anchors(const int target_height,
   {
     feature_maps.push_back(
         {
-            (int) std::ceilf((float) target_height / (float) step),
-            (int) std::ceilf((float) target_width / (float) step)
+            (int) std::ceil((float) target_height / (float) step),
+            (int) std::ceil((float) target_width / (float) step)
         } // ceil
     );
   }
@@ -166,8 +166,8 @@ void FaceBoxes::generate_bboxes(std::vector<types::Boxf> &bbox_collection,
     // ref: https://github.com/zisianw/FaceBoxes.PyTorch/blob/master/utils/box_utils.py
     float cx = prior_cx + dx * variance[0] * prior_s_kx;
     float cy = prior_cy + dy * variance[0] * prior_s_ky;
-    float w = prior_s_kx * std::expf(dw * variance[1]);
-    float h = prior_s_ky * std::expf(dh * variance[1]); // norm coor (0.,1.)
+    float w = prior_s_kx * std::exp(dw * variance[1]);
+    float h = prior_s_ky * std::exp(dh * variance[1]); // norm coor (0.,1.)
 
     types::Boxf box;
     box.x1 = (cx - w / 2.f) * img_width;

lite/ort/cv/mobilenetv2_68.cpp

@@ -45,8 +45,8 @@ void MobileNetV268::detect(const cv::Mat &mat, types::Landmarks &landmarks)
     float x = landmarks_norm.At<float>({0, i});
     float y = landmarks_norm.At<float>({0, i + 1});
 
-    x = std::fmin(std::fmax(0., x), 1.0f);
-    y = std::fmin(std::fmax(0., y), 1.0f);
+    x = std::min(std::max(0.f, x), 1.0f);
+    y = std::min(std::max(0.f, y), 1.0f);
 
     landmarks.points.push_back(cv::Point2f(x * img_width, y * img_height));
   }

lite/ort/cv/mobilenetv2_se_68.cpp

@@ -45,8 +45,8 @@ void MobileNetV2SE68::detect(const cv::Mat &mat, types::Landmarks &landmarks)
     float x = landmarks_norm.At<float>({0, i});
     float y = landmarks_norm.At<float>({0, i + 1});
 
-    x = std::fmin(std::fmax(0., x), 1.0f);
-    y = std::fmin(std::fmax(0., y), 1.0f);
+    x = std::min(std::max(0.f, x), 1.0f);
+    y = std::min(std::max(0.f, y), 1.0f);
 
     landmarks.points.push_back(cv::Point2f(x * img_width, y * img_height));
   }

lite/ort/cv/nanodet.cpp

@@ -21,7 +21,7 @@ void NanoDet::resize_unscale(const cv::Mat &mat, cv::Mat &mat_rs,
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -197,10 +197,10 @@ void NanoDet::generate_bboxes_single_stride(const NanoScaleParams &scale_params,
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/ort/cv/nanodet_efficientnet_lite.cpp

@@ -20,7 +20,7 @@ void NanoDetEfficientNetLite::resize_unscale(const cv::Mat &mat, cv::Mat &mat_rs
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor
@@ -197,10 +197,10 @@ void NanoDetEfficientNetLite::generate_bboxes_single_stride(const NanoLiteScaleP
     float y1 = ((cy - t) * s - (float) dh) / ratio;  // cy - t y1
     float x2 = ((cx + r) * s - (float) dw) / ratio;  // cx + r x2
     float y2 = ((cy + b) * s - (float) dh) / ratio;  // cy + b y2
-    box.x1 = std::fmax(0.f, x1);
-    box.y1 = std::fmax(0.f, y1);
-    box.x2 = std::fmin(img_width, x2);
-    box.y2 = std::fmin(img_height, y2);
+    box.x1 = std::max(0.f, x1);
+    box.y1 = std::max(0.f, y1);
+    box.x2 = std::min(img_width, x2);
+    box.y2 = std::min(img_height, y2);
     box.score = cls_conf;
     box.label = label;
     box.label_text = class_names[label];

lite/ort/cv/pfld.cpp

@@ -44,8 +44,8 @@ void PFLD::detect(const cv::Mat &mat, types::Landmarks &landmarks)
     float x = landmarks_norm.At<float>({0, i});
     float y = landmarks_norm.At<float>({0, i + 1});
 
-    x = std::fmin(std::fmax(0., x), 1.0f);
-    y = std::fmin(std::fmax(0., y), 1.0f);
+    x = std::min(std::max(0.f, x), 1.0f);
+    y = std::min(std::max(0.f, y), 1.0f);
 
     landmarks.points.push_back(cv::Point2f(x * img_width, y * img_height));
   }

lite/ort/cv/pfld68.cpp

@@ -45,8 +45,8 @@ void PFLD68::detect(const cv::Mat &mat, types::Landmarks &landmarks)
     float x = landmarks_norm.At<float>({0, i});
     float y = landmarks_norm.At<float>({0, i + 1});
 
-    x = std::fmin(std::fmax(0., x), 1.0f);
-    y = std::fmin(std::fmax(0., y), 1.0f);
+    x = std::min(std::max(0.f, x), 1.0f);
+    y = std::min(std::max(0.f, y), 1.0f);
 
     landmarks.points.push_back(cv::Point2f(x * img_width, y * img_height));
   }

lite/ort/cv/pfld98.cpp

@@ -44,8 +44,8 @@ void PFLD98::detect(const cv::Mat &mat, types::Landmarks &landmarks)
     float x = landmarks_norm.At<float>({0, i});
     float y = landmarks_norm.At<float>({0, i + 1});
 
-    x = std::fmin(std::fmax(0., x), 1.0f);
-    y = std::fmin(std::fmax(0., y), 1.0f);
+    x = std::min(std::max(0.f, x), 1.0f);
+    y = std::min(std::max(0.f, y), 1.0f);
 
     landmarks.points.push_back(cv::Point2f(x * img_width, y * img_height));
   }

lite/ort/cv/retinaface.cpp

@@ -54,8 +54,8 @@ void RetinaFace::generate_anchors(const int target_height,
   {
     feature_maps.push_back(
         {
-            (int) std::ceilf((float) target_height / (float) step),
-            (int) std::ceilf((float) target_width / (float) step)
+            (int) std::ceil((float) target_height / (float) step),
+            (int) std::ceil((float) target_width / (float) step)
         } // ceil
     );
   }
@@ -131,8 +131,8 @@ void RetinaFace::generate_bboxes(std::vector<types::Boxf> &bbox_collection,
     // ref: https://github.com/biubug6/Pytorch_Retinaface/blob/master/utils/box_utils.py
     float cx = prior_cx + dx * variance[0] * prior_s_kx;
     float cy = prior_cy + dy * variance[0] * prior_s_ky;
-    float w = prior_s_kx * std::expf(dw * variance[1]);
-    float h = prior_s_ky * std::expf(dh * variance[1]); // norm coor (0.,1.)
+    float w = prior_s_kx * std::exp(dw * variance[1]);
+    float h = prior_s_ky * std::exp(dh * variance[1]); // norm coor (0.,1.)
 
     types::Boxf box;
     box.x1 = (cx - w / 2.f) * img_width;

lite/ort/cv/tiny_yolov3.cpp

@@ -88,7 +88,7 @@ std::vector<Ort::Value> TinyYoloV3::transform(const std::vector<cv::Mat> &mats)
   const unsigned int image_height = canvas.rows;
   const unsigned int image_width = canvas.cols;
 
-  const float scale = std::fmin(
+  const float scale = std::min(
       (float) input_width / (float) image_width,
       (float) input_height / (float) image_height
   );

lite/ort/cv/yolop.cpp

@@ -20,7 +20,7 @@ void YOLOP::resize_unscale(const cv::Mat &mat, cv::Mat &mat_rs,
   // scale ratio (new / old) new_shape(h,w)
   float w_r = (float) target_width / (float) img_width;
   float h_r = (float) target_height / (float) img_height;
-  float r = std::fmin(w_r, h_r);
+  float r = std::min(w_r, h_r);
   // compute padding
   int new_unpad_w = static_cast<int>((float) img_width * r); // floor
   int new_unpad_h = static_cast<int>((float) img_height * r); // floor

lite/ort/cv/yolov3.cpp

@@ -88,7 +88,7 @@ std::vector<Ort::Value> YoloV3::transform(const std::vector<cv::Mat> &mats)
   const unsigned int image_height = canvas.rows;
   const unsigned int image_width = canvas.cols;
 
-  const float scale = std::fmin(
+  const float scale = std::min(
       (float) input_width / (float) image_width,
       (float) input_height / (float) image_height
   );