!10220 net: modify num_parallel_worker config in map function to improve performance

From: @ms_yan
Reviewed-by: @heleiwang,@liucunwei
Signed-off-by: @liucunwei

model_zoo/official/cv/googlenet/src/dataset.py

@@ -115,8 +115,8 @@ def create_dataset_imagenet(dataset_path, repeat_num=1, training=True,
 
     transform_label = [C.TypeCast(mstype.int32)]
 
-    data_set = data_set.map(input_columns="image", num_parallel_workers=8, operations=transform_img)
-    data_set = data_set.map(input_columns="label", num_parallel_workers=8, operations=transform_label)
+    data_set = data_set.map(input_columns="image", num_parallel_workers=12, operations=transform_img)
+    data_set = data_set.map(input_columns="label", num_parallel_workers=4, operations=transform_label)
 
     # apply batch operations
     data_set = data_set.batch(imagenet_cfg.batch_size, drop_remainder=True)