diff --git a/.gitignore b/.gitignore
index 38fe8e6afc0c270fd27d62d8ba9106221de9d320..94f052b12456890b0055059f55415f0f1837d733 100644
--- a/.gitignore
+++ b/.gitignore
@@ -14,3 +14,4 @@ model/utils/build/
 imgs/
 *.png
 *.jpg
+misc/
diff --git a/config.py b/config.py
index 8c3c2e6275ebea428290dd379c3f9426d69e8b65..2b5da9949019bdef7d2ad9565e867d4624634806 100644
--- a/config.py
+++ b/config.py
@@ -42,6 +42,7 @@ class Config:
 
     use_adam = False
     use_chainer = False
+    use_drop = False
     # debug
     debug_file = '/tmp/debugf'
 
diff --git a/model/faster_rcnn_vgg16.py b/model/faster_rcnn_vgg16.py
index 73e5e81982b19a46959665276a538ef46ec56fb4..8695c709dfb88089267091237227e012b9e4063c 100644
--- a/model/faster_rcnn_vgg16.py
+++ b/model/faster_rcnn_vgg16.py
@@ -15,9 +15,14 @@ def decom_vgg16(pretrained=True):
     model = vgg16(pretrained)
     features = list(model.features)[:30]
     classifier = model.classifier
-    # classifier = list(classifier)
-    # del the last layer
-    del classifier._modules['6']
+
+    classifier = list(classifier)
+    # delete dropout
+    del classifier[6]
+    if not opt.use_drop:
+        del classifier[5]
+        del classifier[2]
+    classifier = nn.Sequential(*classifier)
 
     # free top3 conv
     for layer in features[:10]:
diff --git a/train.py b/train.py
index d7bf82b26418753e9c6b0bf7cb728c8f81f2cd5e..961923c911c30455d9548fbf8f58e214c77d5a3f 100644
--- a/train.py
+++ b/train.py
@@ -62,6 +62,7 @@ def train(**kwargs):
         trainer.load(opt.load_path)
         print('load pretrained model from %s' % opt.load_path)
 
+    trainer.optimizer = trainer.faster_rcnn.get_great_optimizer()
     trainer.vis.text(dataset.db.label_names, win='labels')
     best_map = 0
     for epoch in range(opt.epoch):
@@ -98,7 +99,7 @@ def train(**kwargs):
                 trainer.vis.text(str(trainer.rpn_cm.value().tolist()), win='rpn_cm')
                 # roi confusion matrix
                 trainer.vis.img('roi_cm', at.totensor(trainer.roi_cm.conf, False).float())
-        if best_map>0.6: 
+        if best_map>0.6 and opt.test_num<5000: 
             opt.test_num=10000
             best_map = 0
         eval_result = eval(test_dataloader, faster_rcnn, test_num=opt.test_num)
@@ -106,9 +107,11 @@ def train(**kwargs):
         if eval_result['map'] > best_map:
             best_map = eval_result['map']
             best_path = trainer.save(best_map=best_map)
-        else:
+        if epoch==8:
             trainer.load(best_path)
             trainer.faster_rcnn.scale_lr(opt.lr_decay)
+        if epoch ==0:
+            trainer.optimizer = trainer.faster_rcnn.get_optimizer()
 
         trainer.vis.plot('test_map', eval_result['map'])
         lr_ = trainer.faster_rcnn.optimizer.param_groups[0]['lr']