diff --git a/ppdet/data/source/dataset.py b/ppdet/data/source/dataset.py
index 9c8e5e647cdeee0df35789ddb89d78d1b96d1a86..3eafbfc3fd4913a8b332c414521ea7b3cf5bb7e1 100644
--- a/ppdet/data/source/dataset.py
+++ b/ppdet/data/source/dataset.py
@@ -203,6 +203,10 @@ class ImageFolder(DetDataset):
         self.image_dir = images
         self.roidbs = self._load_images()
 
+    def get_label_list(self):
+        # Only VOC dataset needs label list in ImageFold 
+        return self.anno_path
+
 
 @register
 class CommonDataset(object):
diff --git a/ppdet/engine/trainer.py b/ppdet/engine/trainer.py
index 38de1f49b48e3e867df9adb33f36d91bbe86bf4e..33c23d6238660abd65cc783e5068b90e5a6d70fe 100644
--- a/ppdet/engine/trainer.py
+++ b/ppdet/engine/trainer.py
@@ -278,12 +278,18 @@ class Trainer(object):
                     save_prediction_only=save_prediction_only)
             ]
         elif self.cfg.metric == 'VOC':
+            output_eval = self.cfg['output_eval'] \
+                if 'output_eval' in self.cfg else None
+            save_prediction_only = self.cfg.get('save_prediction_only', False)
+
             self._metrics = [
                 VOCMetric(
                     label_list=self.dataset.get_label_list(),
                     class_num=self.cfg.num_classes,
                     map_type=self.cfg.map_type,
-                    classwise=classwise)
+                    classwise=classwise,
+                    output_eval=output_eval,
+                    save_prediction_only=save_prediction_only)
             ]
         elif self.cfg.metric == 'WiderFace':
             multi_scale = self.cfg.multi_scale_eval if 'multi_scale_eval' in self.cfg else True
diff --git a/ppdet/metrics/metrics.py b/ppdet/metrics/metrics.py
index 270071b49499bfe8012b5640179b0eaf80310745..b20a569a0434ed7bac7f461399cb280f08ff888a 100644
--- a/ppdet/metrics/metrics.py
+++ b/ppdet/metrics/metrics.py
@@ -225,7 +225,9 @@ class VOCMetric(Metric):
                  map_type='11point',
                  is_bbox_normalized=False,
                  evaluate_difficult=False,
-                 classwise=False):
+                 classwise=False,
+                 output_eval=None,
+                 save_prediction_only=False):
         assert os.path.isfile(label_list), \
                 "label_list {} not a file".format(label_list)
         self.clsid2catid, self.catid2name = get_categories('VOC', label_list)
@@ -233,6 +235,8 @@ class VOCMetric(Metric):
         self.overlap_thresh = overlap_thresh
         self.map_type = map_type
         self.evaluate_difficult = evaluate_difficult
+        self.output_eval = output_eval
+        self.save_prediction_only = save_prediction_only
         self.detection_map = DetectionMAP(
             class_num=class_num,
             overlap_thresh=overlap_thresh,
@@ -245,6 +249,7 @@ class VOCMetric(Metric):
         self.reset()
 
     def reset(self):
+        self.results = {'bbox': [], 'score': [], 'label': []}
         self.detection_map.reset()
 
     def update(self, inputs, outputs):
@@ -256,8 +261,15 @@ class VOCMetric(Metric):
         bbox_lengths = outputs['bbox_num'].numpy() if isinstance(
             outputs['bbox_num'], paddle.Tensor) else outputs['bbox_num']
 
+        self.results['bbox'].append(bboxes.tolist())
+        self.results['score'].append(scores.tolist())
+        self.results['label'].append(labels.tolist())
+
         if bboxes.shape == (1, 1) or bboxes is None:
             return
+        if self.save_prediction_only:
+            return
+
         gt_boxes = inputs['gt_bbox']
         gt_labels = inputs['gt_class']
         difficults = inputs['difficult'] if not self.evaluate_difficult \
@@ -294,6 +306,15 @@ class VOCMetric(Metric):
             bbox_idx += bbox_num
 
     def accumulate(self):
+        output = "bbox.json"
+        if self.output_eval:
+            output = os.path.join(self.output_eval, output)
+            with open(output, 'w') as f:
+                json.dump(self.results, f)
+                logger.info('The bbox result is saved to bbox.json.')
+        if self.save_prediction_only:
+            return
+
         logger.info("Accumulating evaluatation results...")
         self.detection_map.accumulate()