Update train.py (#1591)

fluid/PaddleCV/image_classification/train.py

@@ -242,7 +242,7 @@ def train(args):
         device_num = subprocess.check_output(['nvidia-smi', '-L']).decode().count('\n')
 
     train_batch_size = args.batch_size / device_num
-    test_batch_size = 8
+    test_batch_size = 16
     if not args.enable_ce:
         train_reader = paddle.batch(
             reader.train(), batch_size=train_batch_size, drop_last=True)