修改测试

ultralytics/engine/predictor.py

@@ -253,6 +253,8 @@ class BasePredictor:
                 # Postprocess
                 with profilers[2]:
                     self.results = self.postprocess(preds, im, im0s)
+                if len(self.results) == 0:
+                    continue
                 self.run_callbacks("on_predict_postprocess_end")
 
                 # Visualize, save, write results

ultralytics/models/yolov10/predict.py

@@ -16,14 +16,14 @@ class YOLOv10DetectionPredictor(DetectionPredictor):
             pass
         else:
             preds = preds.transpose(-1, -2)
-            bboxes, scores, labels = ops.v10postprocess(preds, self.args.max_det, preds.shape[-1]-4)
+            bboxes, scores, labels = ops.v10postprocess(preds, self.args.max_det, preds.shape[-1] - 4)
             bboxes = ops.xywh2xyxy(bboxes)
             preds = torch.cat([bboxes, scores.unsqueeze(-1), labels.unsqueeze(-1)], dim=-1)
 
         mask = preds[..., 4] > self.args.conf
         if self.args.classes is not None:
             mask = mask & (preds[..., 5:6] == torch.tensor(self.args.classes, device=preds.device).unsqueeze(0)).any(2)
-        
+
         preds = [p[mask[idx]] for idx, p in enumerate(preds)]
 
         if not isinstance(orig_imgs, list):  # input images are a torch.Tensor, not a list
@@ -31,6 +31,18 @@ class YOLOv10DetectionPredictor(DetectionPredictor):
 
         results = []
         for i, pred in enumerate(preds):
+            #### 不保存负样本predict的结果#######
+            # if pred.numel() == 0:
+            #     continue
+            # print('pred >>> {}'.format(pred[:, 4]))
+            # if float(pred[:, 4][0]) < 0.1:
+            #     continue
+            ##################################
+            #####保存正样本predict漏检的结果######
+            # if pred.numel() != 0:
+            #     if float(pred[:, 4][-1]) > 0.1:
+            #         continue
+            ##################################
             orig_img = orig_imgs[i]
             pred[:, :4] = ops.scale_boxes(img.shape[2:], pred[:, :4], orig_img.shape)
             img_path = self.batch[0][i]

ultralytics/utils/show_trace_pr.py

@@ -26,6 +26,17 @@ class ShowPR:
             values.append(value)
         return values
 
+    def calculate_mena(self, ratio=0.5):
+        values = []
+        for data in self.prec_value:
+            thres_num = int(len(data)*ratio)
+            sorted_data = sorted(data, reverse=True)
+            value = sorted_data[:thres_num]
+            if len(value) == 0:
+                value = sorted_data[:1]
+            values.append(sum(value)/len(value))
+        return values
+
     def _calculate_pr(self, prec_value):
         FN, FP, TN, TP = 0, 0, 0, 0
         for output, target in zip(prec_value, self.tags):
@@ -53,7 +64,7 @@ class ShowPR:
             # print("TP>>{}, FP>>{}, TN>>{}, FN>>{}".format(TP, FP, TN, FN))
         return prec, recall, tn_prec, tn_recall
 
-    def calculate_multiple(self, ratio=0.2):
+    def calculate_multiple_1(self, ratio=0.2):  # 方案1 计算满足阈值判断的占比(ratio)
         recall, recall_TN, PrecisePos, PreciseNeg = [], [], [], []
         for thre in self.thres:
             prec_value = []
@@ -73,6 +84,21 @@ class ShowPR:
             recall_TN.append(tn_recall)
         return recall, recall_TN, PrecisePos, PreciseNeg
 
+    def calculate_multiple_2(self, ratio=0.2):  # 方案2 计算前ratio的预测试值的平均值大于thre为赠品小于为非赠品
+        recall, recall_TN, PrecisePos, PreciseNeg = [], [], [], []
+        event_value = self.calculate_mena(ratio)
+        for thre in self.thres:
+            prec_value = [1 if num >= thre else 0 for num in event_value]
+            prec, recall_pos, tn_prec, tn_recall = self._calculate_pr(prec_value)
+            print(
+                f"thre>>{ratio:.2f},  recall>>{recall_pos:.4f},  precise_pos>>{prec:.4f},  recall_tn>>{tn_recall:.4f},  precise_neg>>{tn_prec:4f}")
+            PrecisePos.append(prec)
+            recall.append(recall_pos)
+            PreciseNeg.append(tn_prec)
+            recall_TN.append(tn_recall)
+        return recall, recall_TN, PrecisePos, PreciseNeg
+
+
     def write_results_to_file(self, recall, recall_TN, PrecisePos, PreciseNeg, ratio):
         file_path = os.sep.join(['./ckpts/tracePR', self.title_name + f"_{ratio:.2f}" + '.txt'])
         with open(file_path, 'w') as file:
@@ -112,5 +138,6 @@ class ShowPR:
             # ratio = 0.5
             if ratio < 0.1 or ratio > 0.95:
                 continue
-            recall, recall_TN, PrecisePos, PreciseNeg = self.calculate_multiple(ratio)
+            recall, recall_TN, PrecisePos, PreciseNeg = self.calculate_multiple_1(ratio)
+            # recall, recall_TN, PrecisePos, PreciseNeg = self.calculate_multiple_2(ratio)
             self.show_pr(recall, recall_TN, PrecisePos, PreciseNeg, ratio)