fix lint

tensorpack/callbacks/inference_runner.py

@@ -25,7 +25,7 @@ from .base import Callback
 from .group import Callbacks
 from .inference import Inferencer
 
-__all__ = ['InferenceRunner',
+__all__ = ['InferenceRunnerBase', 'InferenceRunner',
            'DataParallelInferenceRunner']
 
 
@@ -170,7 +170,9 @@ class DataParallelInferenceRunner(InferenceRunnerBase):
     """
     Inference with data-parallel support on multiple GPUs.
     It will build one predict tower on each GPU, and run prediction
-    with a large total batch.
+    with a large total batch in parallel on all GPUs.
+    It will run the remainder (when the total size of input is not a multiple of #GPU)
+    sequentially.
     """
     def __init__(self, input, infs, gpus):
         """
@@ -188,6 +190,9 @@ class DataParallelInferenceRunner(InferenceRunnerBase):
         assert self._size > 0, "Input for DataParallelInferenceRunner must have a size!"
         self._gpus = gpus
 
+        self._hooks = []
+        self._hooks_parallel = []
+
     def _setup_graph(self):
         self._handles = []
 
@@ -209,15 +214,18 @@ class DataParallelInferenceRunner(InferenceRunnerBase):
         # e.g. hooks from StagingInput will force the consumption
         # of nr_tower datapoints in every run.
         input_hooks = self._input_callbacks.get_hooks()
-        self._hooks = [self._build_hook(inf) for inf in self.infs] + input_hooks
-        self._hooks_parallel = [self._build_hook_parallel(inf) for inf in self.infs] + input_hooks
+        self._hooks.extend([self._build_hook(inf) for inf in self.infs] + input_hooks)
+        self._hooks_parallel.extend([self._build_hook_parallel(inf) for inf in self.infs] + input_hooks)
 
         for inf in self.infs:
             inf.setup_graph(self.trainer)
         self._input_callbacks.setup_graph(self.trainer)
 
     def register_hook(self, h):
-        raise NotImplementedError("DataParallelInferenceRunner doesn't accept extra hooks!")
+        logger.info(
+            "[DataParallelInferenceRunner] Registering hook {} on both parallel and sequential inference.")
+        self._hooks.append(h)
+        self._hooks_parallel.append(h)
 
     class InferencerToHookDataParallel(InferencerToHook):
         def __init__(self, inf, fetches, size):

tensorpack/graph_builder/utils.py

@@ -7,7 +7,7 @@ from contextlib import contextmanager
 import operator
 import tensorflow as tf
 
-from ..tfutils.common  import get_tf_version_number
+from ..tfutils.common import get_tf_version_number
 
 
 __all__ = ['LeastLoadedDeviceSetter',
@@ -45,18 +45,17 @@ def override_to_local_variable(enable=True):
         orig_vs = tf.get_variable_scope()
         if get_tf_version_number() >= 1.5:
             with tf.variable_scope(
-                tf.get_variable_scope(),
-                custom_getter=custom_getter,
-                auxiliary_name_scope=False):
+                    orig_vs,
+                    custom_getter=custom_getter,
+                    auxiliary_name_scope=False):
                 yield
         else:
             if get_tf_version_number() >= 1.2:
                 ns = tf.get_default_graph().get_name_scope()
             else:
-                ns = tf.get_variable_scope().original_name_scope
+                ns = orig_vs.original_name_scope
             with tf.variable_scope(
-                    tf.get_variable_scope(),
-                    custom_getter=custom_getter):
+                    orig_vs, custom_getter=custom_getter):
                 with tf.name_scope(ns + '/'):
                     yield
     else:

tensorpack/input_source/input_source.py

@@ -121,7 +121,7 @@ class EnqueueThread(ShareSessionThread):
         # self._size = queue.size()
 
     def run(self):
-        with self.default_sess() as sess:
+        with self.default_sess():
             try:
                 self.reinitialize_dataflow()
                 while True: