boost NMS performance

diff --git a/modules/dnn/src/layers/detection_output_layer.cpp b/modules/dnn/src/layers/detection_output_layer.cpp
index 2dd6f5fb73bc32db0ba67d87a13b5c4fb60d30a5..8374d74293bee05f84f80ada5f8a9ec0a1cb2fd9 100644 (file)
--- a/modules/dnn/src/layers/detection_output_layer.cpp
+++ b/modules/dnn/src/layers/detection_output_layer.cpp
@@ -133,6 +133,12 @@ public:
 
     typedef std::map<int, std::vector<util::NormalizedBBox> > LabelBBox;
 
+    inline int getNumOfTargetClasses() {
+        unsigned numBackground =
+            (_backgroundLabelId >= 0 && _backgroundLabelId < _numClasses) ? 1 : 0;
+        return (_numClasses - numBackground);
+    }
+
     bool getParameterDict(const LayerParams &params,
                           const std::string &parameterName,
                           DictValue& result)
@@ -584,12 +590,13 @@ public:
             LabelBBox::const_iterator label_bboxes = decodeBBoxes.find(label);
             if (label_bboxes == decodeBBoxes.end())
                 CV_Error_(cv::Error::StsError, ("Could not find location predictions for label %d", label));
+            int limit = (getNumOfTargetClasses() == 1) ? _keepTopK : std::numeric_limits<int>::max();
             if (_bboxesNormalized)
                 NMSFast_(label_bboxes->second, scores, _confidenceThreshold, _nmsThreshold, 1.0, _topK,
-                         indices[c], util::caffe_norm_box_overlap);
+                         indices[c], util::caffe_norm_box_overlap, limit);
             else
                 NMSFast_(label_bboxes->second, scores, _confidenceThreshold, _nmsThreshold, 1.0, _topK,
-                         indices[c], util::caffe_box_overlap);
+                         indices[c], util::caffe_box_overlap, limit);
             numDetections += indices[c].size();
         }
         if (_keepTopK > -1 && numDetections > (size_t)_keepTopK)

diff --git a/modules/dnn/src/nms.inl.hpp b/modules/dnn/src/nms.inl.hpp
index 89e3adfcf57e2bf81e06e75e23652cd02821e208..7b84839c02c6550534526f3bc696a58dbec463d6 100644 (file)
--- a/modules/dnn/src/nms.inl.hpp
+++ b/modules/dnn/src/nms.inl.hpp
@@ -62,12 +62,15 @@ inline void GetMaxScoreIndex(const std::vector<float>& scores, const float thres
 //    score_threshold: a threshold used to filter detection results.
 //    nms_threshold: a threshold used in non maximum suppression.
 //    top_k: if not > 0, keep at most top_k picked indices.
+//    limit: early terminate once the # of picked indices has reached it.
 //    indices: the kept indices of bboxes after nms.
 template <typename BoxType>
 inline void NMSFast_(const std::vector<BoxType>& bboxes,
       const std::vector<float>& scores, const float score_threshold,
       const float nms_threshold, const float eta, const int top_k,
-      std::vector<int>& indices, float (*computeOverlap)(const BoxType&, const BoxType&))
+      std::vector<int>& indices,
+      float (*computeOverlap)(const BoxType&, const BoxType&),
+      int limit = std::numeric_limits<int>::max())
 {
     CV_Assert(bboxes.size() == scores.size());
 
@@ -86,8 +89,12 @@ inline void NMSFast_(const std::vector<BoxType>& bboxes,
             float overlap = computeOverlap(bboxes[idx], bboxes[kept_idx]);
             keep = overlap <= adaptive_threshold;
         }
-        if (keep)
+        if (keep) {
             indices.push_back(idx);
+            if (indices.size() >= limit) {
+                break;
+            }
+        }
         if (keep && eta < 1 && adaptive_threshold > 0.5) {
           adaptive_threshold *= eta;
         }