tf_upgrade_v2 for cmd

blueoil/cmd/convert_weight_from_darknet.py

@@ -53,14 +53,14 @@ def convert(config, weight_file):
 
         model.inference(images_placeholder, is_training)
 
-        init_op = tf.global_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
 
         saver = tf.compat.v1.train.Saver(max_to_keep=None)
 
-        variables = tf.global_variables()
+        variables = tf.compat.v1.global_variables()
 
     session_config = None
-    sess = tf.Session(graph=graph, config=session_config)
+    sess = tf.compat.v1.Session(graph=graph, config=session_config)
     sess.run([init_op, ])
     suffixes = ['bias', 'beta', 'gamma', 'moving_mean', 'moving_variance', 'kernel']
     convert_variables = []

blueoil/cmd/evaluate.py

@@ -106,15 +106,15 @@ def evaluate(config, restore_path, output_dir):
 
         metrics_summary_op, metrics_placeholders = executor.prepare_metrics(metrics_ops_dict)
 
-        init_op = tf.global_variables_initializer()
-        reset_metrics_op = tf.local_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
+        reset_metrics_op = tf.compat.v1.local_variables_initializer()
         saver = tf.compat.v1.train.Saver(max_to_keep=None)
 
     session_config = None  # tf.ConfigProto(log_device_placement=True)
-    sess = tf.Session(graph=graph, config=session_config)
+    sess = tf.compat.v1.Session(graph=graph, config=session_config)
     sess.run([init_op, reset_metrics_op])
 
-    validation_writer = tf.summary.FileWriter(environment.TENSORBOARD_DIR + "/evaluate")
+    validation_writer = tf.compat.v1.summary.FileWriter(environment.TENSORBOARD_DIR + "/evaluate")
 
     saver.restore(sess, restore_path)
 

blueoil/cmd/export.py

@@ -95,12 +95,12 @@ def _export(config, restore_path, image_path):
 
         images_placeholder, _ = model.placeholders()
         model.inference(images_placeholder, is_training)
-        init_op = tf.global_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
 
         saver = tf.compat.v1.train.Saver(max_to_keep=50)
 
-    session_config = tf.ConfigProto()
-    sess = tf.Session(graph=graph, config=session_config)
+    session_config = tf.compat.v1.ConfigProto()
+    sess = tf.compat.v1.Session(graph=graph, config=session_config)
     sess.run(init_op)
 
     saver.restore(sess, restore_path)

blueoil/cmd/measure_latency.py

@@ -55,13 +55,13 @@ def _measure_time(config, restore_path, step_size):
         images_placeholder, labels_placeholder = model.placeholders()
         output = model.inference(images_placeholder, is_training)
 
-        init_op = tf.global_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
 
         saver = tf.compat.v1.train.Saver()
 
     session_config = None  # tf.ConfigProto(log_device_placement=True)
     # session_config.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1
-    sess = tf.Session(graph=graph, config=session_config)
+    sess = tf.compat.v1.Session(graph=graph, config=session_config)
     sess.run(init_op)
 
     if restore_path:

blueoil/cmd/predict.py

@@ -77,12 +77,12 @@ def _run(input_dir, output_dir, config, restore_path, save_images):
         images_placeholder, _ = model.placeholders()
         output_op = model.inference(images_placeholder, is_training)
 
-        init_op = tf.global_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
 
         saver = tf.compat.v1.train.Saver(max_to_keep=None)
 
-    session_config = tf.ConfigProto()
-    sess = tf.Session(graph=graph, config=session_config)
+    session_config = tf.compat.v1.ConfigProto()
+    sess = tf.compat.v1.Session(graph=graph, config=session_config)
     sess.run(init_op)
     saver.restore(sess, restore_path)
 

blueoil/cmd/profile_model.py

@@ -54,11 +54,11 @@ def _profile(config, restore_path, bit, unquant_layers):
         images_placeholder, _ = model.placeholders()
         model.inference(images_placeholder, is_training)
 
-        init_op = tf.global_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
         saver = tf.compat.v1.train.Saver(max_to_keep=50)
 
-    session_config = tf.ConfigProto()
-    sess = tf.Session(graph=graph, config=session_config)
+    session_config = tf.compat.v1.ConfigProto()
+    sess = tf.compat.v1.Session(graph=graph, config=session_config)
     sess.run(init_op)
 
     if restore_path:
@@ -145,7 +145,8 @@ def _save_json(name, image_size, num_classes, node_param_dict, node_flops_dict):
 
 
 def _profile_flops(graph, res, scopes):
-    float_prof = tf.profiler.profile(graph, options=tf.profiler.ProfileOptionBuilder.float_operation())
+    float_prof = tf.compat.v1.profiler.profile(
+        graph, options=tf.compat.v1.profiler.ProfileOptionBuilder.float_operation())
     float_res_dict = collections.defaultdict(int)
     float_res_dict["total"] = float_prof.total_float_ops
     for node in float_prof.children:
@@ -172,7 +173,8 @@ def _profile_flops(graph, res, scopes):
 
 
 def _profile_params(graph, res, bit, unquant_layers):
-    prof = tf.profiler.profile(graph, options=tf.profiler.ProfileOptionBuilder.trainable_variables_parameter())
+    prof = tf.compat.v1.profiler.profile(
+        graph, options=tf.compat.v1.profiler.ProfileOptionBuilder.trainable_variables_parameter())
 
     # helper func to make profile res
     def helper(node, level):

blueoil/cmd/train.py

@@ -129,8 +129,8 @@ def start_training(config):
 
         metrics_summary_op, metrics_placeholders = executor.prepare_metrics(metrics_ops_dict)
 
-        init_op = tf.global_variables_initializer()
-        reset_metrics_op = tf.local_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
+        reset_metrics_op = tf.compat.v1.local_variables_initializer()
         if use_horovod:
             # add Horovod broadcasting variables from rank 0 to all
             bcast_global_variables_op = hvd.broadcast_global_variables(0)
@@ -141,7 +141,7 @@ def start_training(config):
             saver = tf.compat.v1.train.Saver(max_to_keep=config.KEEP_CHECKPOINT_MAX)
 
         if config.IS_PRETRAIN:
-            all_vars = tf.global_variables()
+            all_vars = tf.compat.v1.global_variables()
             pretrain_var_list = [
                 var for var in all_vars if var.name.startswith(tuple(config.PRETRAIN_VARS))
             ]
@@ -152,8 +152,8 @@ def start_training(config):
 
     if use_horovod:
         # For distributed training
-        session_config = tf.ConfigProto(
-            gpu_options=tf.GPUOptions(
+        session_config = tf.compat.v1.ConfigProto(
+            gpu_options=tf.compat.v1.GPUOptions(
                 allow_growth=True,
                 visible_device_list=str(hvd.local_rank())
             )
@@ -166,23 +166,24 @@ def start_training(config):
         #         per_process_gpu_memory_fraction=0.1
         #     )
         # )
-        session_config = tf.ConfigProto()  # tf.ConfigProto(log_device_placement=True)
+        session_config = tf.compat.v1.ConfigProto()  # tf.ConfigProto(log_device_placement=True)
     # TODO(wakisaka): XLA JIT
     # session_config.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1
 
-    sess = tf.Session(graph=graph, config=session_config)
+    sess = tf.compat.v1.Session(graph=graph, config=session_config)
     sess.run([init_op, reset_metrics_op])
 
     if rank == 0:
-        train_writer = tf.summary.FileWriter(environment.TENSORBOARD_DIR + "/train", sess.graph)
+        train_writer = tf.compat.v1.summary.FileWriter(environment.TENSORBOARD_DIR + "/train", sess.graph)
         if use_train_validation_saving:
-            train_val_saving_writer = tf.summary.FileWriter(environment.TENSORBOARD_DIR + "/train_validation_saving")
-        val_writer = tf.summary.FileWriter(environment.TENSORBOARD_DIR + "/validation")
+            train_val_saving_writer = tf.compat.v1.summary.FileWriter(
+                environment.TENSORBOARD_DIR + "/train_validation_saving")
+        val_writer = tf.compat.v1.summary.FileWriter(environment.TENSORBOARD_DIR + "/validation")
 
         if config.IS_PRETRAIN:
             print("------- Load pretrain data ----------")
             pretrain_saver.restore(sess, os.path.join(config.PRETRAIN_DIR, config.PRETRAIN_FILE))
-            sess.run(tf.assign(global_step, 0))
+            sess.run(tf.compat.v1.assign(global_step, 0))
 
         last_step = 0
 
@@ -301,7 +302,7 @@ def start_training(config):
 
             if step == 0:
                 # check create pb on only first step.
-                minimal_graph = tf.graph_util.convert_variables_to_constants(
+                minimal_graph = tf.compat.v1.graph_util.convert_variables_to_constants(
                     sess,
                     sess.graph.as_graph_def(add_shapes=True),
                     ["output"],

blueoil/cmd/tune_ray.py

@@ -116,14 +116,14 @@ def update_parameters_for_each_trial(network_kwargs, chosen_kwargs):
             network_kwargs['optimizer_kwargs'][key] = chosen_kwargs['optimizer_class'][key]
     network_kwargs['learning_rate_func'] = chosen_kwargs['learning_rate_func']['scheduler']
     base_lr = chosen_kwargs['learning_rate']
-    if network_kwargs['learning_rate_func'] is tf.train.piecewise_constant:
+    if network_kwargs['learning_rate_func'] is tf.compat.v1.train.piecewise_constant:
         lr_factor = chosen_kwargs['learning_rate_func']['scheduler_factor']
         network_kwargs['learning_rate_kwargs']['values'] = [base_lr,
                                                             base_lr * lr_factor,
                                                             base_lr * lr_factor * lr_factor,
                                                             base_lr * lr_factor * lr_factor * lr_factor]
         network_kwargs['learning_rate_kwargs']['boundaries'] = chosen_kwargs['learning_rate_func']['scheduler_steps']
-    elif network_kwargs['learning_rate_func'] is tf.train.polynomial_decay:
+    elif network_kwargs['learning_rate_func'] is tf.compat.v1.train.polynomial_decay:
         network_kwargs['learning_rate_kwargs']['learning_rate'] = base_lr
         network_kwargs['learning_rate_kwargs']['power'] = chosen_kwargs['learning_rate_func']['scheduler_power']
         network_kwargs['learning_rate_kwargs']['decay_steps'] = chosen_kwargs['learning_rate_func']['scheduler_decay']
@@ -210,12 +210,12 @@ def _setup(self, config):
         self.metrics_ops_dict = metrics_ops_dict
         self.metrics_update_op = metrics_update_op
 
-        init_op = tf.global_variables_initializer()
-        self.reset_metrics_op = tf.local_variables_initializer()
+        init_op = tf.compat.v1.global_variables_initializer()
+        self.reset_metrics_op = tf.compat.v1.local_variables_initializer()
 
-        session_config = tf.ConfigProto(
-            gpu_options=tf.GPUOptions(allow_growth=True))
-        self.sess = tf.Session(config=session_config)
+        session_config = tf.compat.v1.ConfigProto(
+            gpu_options=tf.compat.v1.GPUOptions(allow_growth=True))
+        self.sess = tf.compat.v1.Session(config=session_config)
         self.sess.run([init_op, self.reset_metrics_op])
         self.iterations = 0
         self.saver = tf.compat.v1.train.Saver()