Add DetectionMAPEvaluator #2467

pkuyym · 2017-06-14T14:20:35Z

resolves #2466

… ssd_map

qingqing01 · 2017-06-16T06:29:34Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+      map<size_t, vector<NormalizedBBox>> bboxes;
+      for (int i = labelIndex[n]; i < labelIndex[n + 1]; ++i) {
+        vector<NormalizedBBox> bbox;
+        getBBoxFromLabelData(cpuLabel_->getData() + i * 6, 1, bbox);


和layout相关6，7常量后续的PR需要重新设计，避免固定死，这里标记下

qingqing01 · 2017-06-16T06:44:49Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+        ++n;
+      }
+      ++imgId;
+      if (imgId > batchSize) break;


什么时候imgId > batchSize?

Rewrite this part.

qingqing01 · 2017-06-16T07:01:59Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+
+    // calcTFPos
+    calcTFPos(
+        batchSize, allGTBBoxes, allDetectBBoxes, &allTruePos_, &allFalsePos_);


allTruePos_，allFalsePos_都已经是成员变量了，这里也不用当做参数传吧

qingqing01 · 2017-06-16T07:21:05Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+
+  map<size_t, size_t> numPos_;
+  map<size_t, vector<pair<real, size_t>>> allTruePos_;
+  map<size_t, vector<pair<real, size_t>>> allFalsePos_;


对变量增加注释~

qingqing01 · 2017-06-16T07:28:07Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+
+  std::string getTypeImpl() const { return "detection_map"; }
+
+  real getValueImpl() const { return calcMAP() * 100; }


100放到calcMAP()里吧，这样130行也不用乘了~

qingqing01 · 2017-06-16T07:30:08Z

python/paddle/trainer_config_helpers/evaluators.py

+                            ap_type="11point",
+                            name=None):
+    """
+    Detection mAP Evaluator. It will print mean Average Precision for detection.


mean Average Precision -> mean Average Precision(mAP)

qingqing01 · 2017-06-16T07:32:14Z

python/paddle/trainer_config_helpers/evaluators.py

+    """
+    Detection mAP Evaluator. It will print mean Average Precision for detection.
+
+    The detection mAP Evaluator according to the detection_output's output count


count-> counts

qingqing01 · 2017-06-16T07:35:53Z

python/paddle/trainer_config_helpers/evaluators.py

+    Detection mAP Evaluator. It will print mean Average Precision for detection.
+
+    The detection mAP Evaluator according to the detection_output's output count
+    the true positive and the false positive bbox and integral them to get the


integral -> integrates

这里是积分哈

qingqing01 · 2017-06-16T07:37:04Z

python/paddle/trainer_config_helpers/evaluators.py

+    """
+    Detection mAP Evaluator. It will print mean Average Precision for detection.
+
+    The detection mAP Evaluator according to the detection_output's output count


according to the detection_output's output

based on the output detection_output layer.

qingqing01 · 2017-06-16T07:37:56Z

python/paddle/trainer_config_helpers/evaluators.py

+    :type overlap_threshold: float
+    :param background_id: The background class index.
+    :type background_id: int
+    :param evaluate_difficult: Wether evaluate a difficult ground truth.


Wether - Whether

pkuyym

Follow comments.

pkuyym · 2017-06-19T08:08:00Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+
+  std::string getTypeImpl() const { return "detection_map"; }
+
+  real getValueImpl() const { return calcMAP() * 100; }


pkuyym · 2017-06-19T08:26:54Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+
+  map<size_t, size_t> numPos_;
+  map<size_t, vector<pair<real, size_t>>> allTruePos_;
+  map<size_t, vector<pair<real, size_t>>> allFalsePos_;


pkuyym · 2017-06-19T08:27:14Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+
+    // calcTFPos
+    calcTFPos(
+        batchSize, allGTBBoxes, allDetectBBoxes, &allTruePos_, &allFalsePos_);


pkuyym · 2017-06-19T08:32:01Z

python/paddle/trainer_config_helpers/evaluators.py

+                            ap_type="11point",
+                            name=None):
+    """
+    Detection mAP Evaluator. It will print mean Average Precision for detection.


pkuyym · 2017-06-19T08:32:15Z

python/paddle/trainer_config_helpers/evaluators.py

+    """
+    Detection mAP Evaluator. It will print mean Average Precision for detection.
+
+    The detection mAP Evaluator according to the detection_output's output count


pkuyym · 2017-06-19T08:33:34Z

python/paddle/trainer_config_helpers/evaluators.py

+    """
+    Detection mAP Evaluator. It will print mean Average Precision for detection.
+
+    The detection mAP Evaluator according to the detection_output's output count


pkuyym · 2017-06-19T08:34:05Z

python/paddle/trainer_config_helpers/evaluators.py

+    Detection mAP Evaluator. It will print mean Average Precision for detection.
+
+    The detection mAP Evaluator according to the detection_output's output count
+    the true positive and the false positive bbox and integral them to get the


这里是积分哈

pkuyym · 2017-06-19T08:34:36Z

python/paddle/trainer_config_helpers/evaluators.py

+    :type overlap_threshold: float
+    :param background_id: The background class index.
+    :type background_id: int
+    :param evaluate_difficult: Wether evaluate a difficult ground truth.


pkuyym · 2017-06-19T09:17:28Z

python/paddle/trainer_config_helpers/evaluators.py

@@ -21,7 +21,8 @@
    "chunk_evaluator", "sum_evaluator", "column_sum_evaluator",
    "value_printer_evaluator", "gradient_printer_evaluator",
    "maxid_printer_evaluator", "maxframe_printer_evaluator",
-    "seqtext_printer_evaluator", "classification_error_printer_evaluator"
+    "seqtext_printer_evaluator", "classification_error_printer_evaluator",
+    "detection_map_evaluator"


pkuyym · 2017-06-19T10:15:38Z

paddle/gserver/evaluators/DetectionMAPEvaluator.cpp

+        ++n;
+      }
+      ++imgId;
+      if (imgId > batchSize) break;


Rewrite this part.

qingqing01

LGTM.

Add DetectionMAPEvaluator.

597a58c

pkuyym requested a review from qingqing01 June 14, 2017 14:20

pkuyym changed the title ~~Add DetectionMAPEvaluator.~~ Add DetectionMAPEvaluator Jun 14, 2017

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

2207659

… ssd_map

qingqing01 requested changes Jun 16, 2017

View reviewed changes

Follow comments.

5f924d5

pkuyym commented Jun 19, 2017

View reviewed changes

qingqing01 approved these changes Jun 20, 2017

View reviewed changes

pkuyym merged commit f551c27 into PaddlePaddle:develop Jun 23, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add DetectionMAPEvaluator #2467

Add DetectionMAPEvaluator #2467

pkuyym commented Jun 14, 2017

qingqing01 Jun 16, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

qingqing01 Jun 16, 2017

pkuyym Jun 19, 2017

pkuyym left a comment

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

pkuyym Jun 19, 2017

qingqing01 left a comment


		std::string getTypeImpl() const { return "detection_map"; }

		real getValueImpl() const { return calcMAP() * 100; }

Add DetectionMAPEvaluator #2467

Add DetectionMAPEvaluator #2467

Conversation

pkuyym commented Jun 14, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

pkuyym left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

qingqing01 left a comment

Choose a reason for hiding this comment