add class accuracy

zdaiot · Sep 27, 2019 · 9601ca6 · 9601ca6
1 parent 0f1803f
commit 9601ca6
Show file tree

Hide file tree

Showing 2 changed files with 31 additions and 12 deletions.
diff --git a/train_classify.py b/train_classify.py
@@ -95,11 +95,14 @@ def train(self, train_loader, valid_loader):
             print('Finish Epoch [%d/%d], Average Loss: %.7f' % (epoch, self.epoch, epoch_loss/len(tbar)))
 
             # 验证模型
-            neg_accuracy, pos_accuracy, accuracy, loss_valid = self.validation(valid_loader)
+            class_neg_accuracy, class_pos_accuracy, class_accuracy, neg_accuracy, pos_accuracy, accuracy, loss_valid = \
+                self.validation(valid_loader)
+
             if accuracy > self.max_accuracy_valid: 
                 is_best = True
                 self.max_accuracy_valid = accuracy
-            else: is_best = False
+            else:
+                is_best = False
 
             state = {
                 'epoch': epoch,
@@ -110,6 +113,10 @@ def train(self, train_loader, valid_loader):
             self.solver.save_checkpoint(os.path.join(self.model_path, '%s_classify_fold%d.pth' % (self.model_name, self.fold)), state, is_best)
             self.writer.add_scalar('valid_loss', loss_valid, epoch)
             self.writer.add_scalar('valid_accuracy', accuracy, epoch)
+            self.writer.add_scalar('valid_class_0_accuracy', class_accuracy[0], epoch)
+            self.writer.add_scalar('valid_class_1_accuracy', class_accuracy[1], epoch)
+            self.writer.add_scalar('valid_class_2_accuracy', class_accuracy[2], epoch)
+            self.writer.add_scalar('valid_class_3_accuracy', class_accuracy[3], epoch)
 
     def validation(self, valid_loader):
         ''' 完成模型的验证过程
@@ -135,9 +142,13 @@ def validation(self, valid_loader):
                 tbar.set_description(desc=descript)
         loss_mean = loss_sum / len(tbar)
 
-        neg_accuracy, pos_accuracy, accuracy = meter.get_metrics()
-        print("Negative accuracy: %0.4f | positive accuracy: %0.4f | accuracy: %0.4f" % (neg_accuracy, pos_accuracy, accuracy))
-        return neg_accuracy, pos_accuracy, accuracy, loss_mean
+        class_neg_accuracy, class_pos_accuracy, class_accuracy, neg_accuracy, pos_accuracy, accuracy = meter.get_metrics()
+        print("Class_0_accuracy: %0.4f | Class_1_accuracy: %0.4f | Class_2_accuracy: %0.4f | Class_3_accuracy: %0.4f | "
+              "Negative accuracy: %0.4f | positive accuracy: %0.4f | accuracy: %0.4f" %
+              (class_accuracy[0], class_accuracy[1], class_accuracy[2], class_accuracy[3],
+               neg_accuracy, pos_accuracy, accuracy))
+        return class_neg_accuracy, class_pos_accuracy, class_accuracy, neg_accuracy, pos_accuracy, accuracy, loss_mean
+
 
 class ChooseMinArea():
     def __init__(self, ):

diff --git a/utils/cal_classify_accuracy.py b/utils/cal_classify_accuracy.py
@@ -28,11 +28,11 @@ def metric(logit, truth, threshold=0.5):
         num_pos = t.sum(dim=[0, 2])
         num_neg = batch_size * H * W - num_pos
         # 预测正确的正样本和负样本的数目
-        tp = tp.data.cpu().numpy().sum()
-        tn = tn.data.cpu().numpy().sum()
+        tp = tp.data.cpu().numpy()
+        tn = tn.data.cpu().numpy()
         # 正样本、负样本的数目
-        num_pos = num_pos.data.cpu().numpy().sum()
-        num_neg = num_neg.data.cpu().numpy().sum()
+        num_pos = num_pos.data.cpu().numpy()
+        num_neg = num_neg.data.cpu().numpy()
 
         # tp = np.nan_to_num(tp / (num_pos + 1e-12), 0)
         # tn = np.nan_to_num(tn / (num_neg + 1e-12), 0)
@@ -60,15 +60,23 @@ def update(self, targets, outputs):
         self.number_positive.append(num_pos)
 
     def get_metrics(self):
-        # 预测正确的样本的数目
+        # 各类预测正确的样本数目，样本总数目
+        class_tn = np.sum(np.array(self.true_negative), axis=0)
+        class_tp = np.sum(np.array(self.true_poisitive), axis=0)
+        class_num_neg = np.sum(np.array(self.number_negative), axis=0)
+        class_num_pos = np.sum(np.array(self.number_positive), axis=0)
+        # 预测正确的样本的总数目，样本总数目
         tn = np.sum(self.true_negative)
         tp = np.sum(self.true_poisitive)
-        # 负样本和正样本各自的数目
         num_neg = np.sum(self.number_negative)
         num_pos = np.sum(self.number_positive)
+        # 各类的正负样本的准确率和总的准确率
+        class_neg_accuracy = class_tn / class_num_neg
+        class_pos_accuracy = class_tp / class_num_pos
+        class_accuracy = (class_tn + class_tp) / (class_num_neg + class_num_pos)
         # 正负样本各自的准确率和总的准确率
         neg_accuracy = tn / (num_neg + 1e-12)
         pos_accuracy = tp / (num_pos + 1e-12)
         accuracy = (tn + tp) / (num_neg + num_pos)
 
-        return neg_accuracy, pos_accuracy, accuracy
+        return class_neg_accuracy, class_pos_accuracy, class_accuracy, neg_accuracy, pos_accuracy, accuracy