Follow comments.

Author: qingqing01

fluid/image_classification/se_resnext.py

@@ -1,6 +1,7 @@
 import paddle.v2 as paddle
 import paddle.fluid as fluid
 
+
 def conv_bn_layer(input, num_filters, filter_size, stride=1, groups=1,
  act=None):
  conv = fluid.layers.conv2d(

fluid/image_classification/train.py

@@ -16,9 +16,11 @@
 # yapf: disable
 add_arg('batch_size', int, 256, "Minibatch size.")
 add_arg('num_layers', int, 50, "How many layers for SE-ResNeXt model.")
-add_arg('parallel_exe', bool, True, "Whether use ParallelExecutor to train.")
+add_arg('with_mem_opt', bool, True, "Whether to use memory optimization or not.")
+add_arg('parallel_exe', bool, True, "Whether to use ParallelExecutor to train or not.")
 
-def train_paralle_do(learning_rate,
+def train_paralle_do(args,
+ learning_rate,
  batch_size,
  num_passes,
  init_model=None,
@@ -74,13 +76,12 @@ def train_paralle_do(learning_rate,
  momentum=0.9,
  regularization=fluid.regularizer.L2Decay(1e-4))
 
- opts = optimizer.minimize(avg_cost)
- fluid.memory_optimize(fluid.default_main_program())
+ inference_program = fluid.default_main_program().clone(for_test=True)
 
- inference_program = fluid.default_main_program().clone()
- with fluid.program_guard(inference_program):
- inference_program = fluid.io.get_inference_program(
-  [avg_cost, acc_top1, acc_top5])
+ opts = optimizer.minimize(avg_cost)
+ if args.with_mem_opt:
+ fluid.memory_optimize(fluid.default_main_program())
+ fluid.memory_optimize(inference_program)
 
  place = fluid.CUDAPlace(0)
  exe = fluid.Executor(place)
@@ -153,7 +154,8 @@ def train_paralle_do(learning_rate,
  os.makedirs(model_path)
  fluid.io.save_persistables(exe, model_path)
 
-def train_parallel_exe(learning_rate,
+def train_parallel_exe(args,
+ learning_rate,
  batch_size,
  num_passes,
  init_model=None,
@@ -190,7 +192,10 @@ def train_parallel_exe(learning_rate,
  regularization=fluid.regularizer.L2Decay(1e-4))
 
  opts = optimizer.minimize(avg_cost)
- fluid.memory_optimize(fluid.default_main_program())
+
+ if args.with_mem_opt:
+ fluid.memory_optimize(fluid.default_main_program())
+ fluid.memory_optimize(test_program)
 
  place = fluid.CUDAPlace(0)
  exe = fluid.Executor(place)
@@ -199,13 +204,8 @@ def train_parallel_exe(learning_rate,
  if init_model is not None:
  fluid.io.load_persistables(exe, init_model)
 
- import paddle.dataset.flowers as flowers
-
- #train_reader = paddle.batch(reader.train(), batch_size=batch_size)
- #test_reader = paddle.batch(reader.test(), batch_size=batch_size)
-
- train_reader = paddle.batch(flowers.train(), batch_size=batch_size)
- test_reader = paddle.batch(flowers.test(), batch_size=batch_size)
+ train_reader = paddle.batch(reader.train(), batch_size=batch_size)
+ test_reader = paddle.batch(reader.test(), batch_size=batch_size)
  feeder = fluid.DataFeeder(place=place, feed_list=[image, label])
 
  train_exe = fluid.ParallelExecutor(use_cuda=True, loss_name=avg_cost.name)
@@ -300,7 +300,8 @@ def train_parallel_exe(learning_rate,
  # layers: 50, 152
  layers = args.num_layers
  method = train_parallel_exe if args.parallel_exe else train_parallel_do
- method(learning_rate=0.1,
+ method(args,
+ learning_rate=0.1,
  batch_size=batch_size,
  num_passes=120,
  init_model=None,