update readme

examples/FasterRCNN/README.md

 # Faster-RCNN on COCO
-This example aimes to provide a minimal Multi-GPU implementation (<1000 lines) of ResNet50-Faster-RCNN on COCO.
+This example aimes to provide a minimal (<1000 lines) Multi-GPU implementation of ResNet50-Faster-RCNN on COCO.
 
 ## Dependencies
 + TensorFlow nightly.
@@ -44,13 +44,13 @@ To predict on an image (and show output in a window):
 The hyperparameters are not carefully tuned. You can probably get better performance by e.g.  training longer.
 
 ## Files
-This is an minimal implementation that simply contains these files:
+This is a minimal implementation that simply contains these files:
 + coco.py: load COCO data
 + data.py: prepare data for training
-+ common.py: some common data preparation utilities
++ common.py: common data preparation utilities
 + basemodel.py: implement resnet
-+ model.py: implement faster-rcnn
-+ viz.py: visualization utilities
-+ utils/: third-party helper functions
++ model.py: implement rpn/faster-rcnn
 + train.py: main training script
-+ eval.py: utilities for evaluation
++ utils/: third-party helper functions
++ eval.py: evaluation utilities
++ viz.py: visualization utilities

examples/FasterRCNN/data.py

@@ -248,7 +248,7 @@ def get_train_dataflow():
 
 def get_eval_dataflow():
     imgs = COCODetection.load_many(config.BASEDIR, config.VAL_DATASET, add_gt=False)
-# no filter for training
+    # no filter for training
     ds = DataFromListOfDict(imgs, ['file_name', 'id'])
 
     def f(fname):

tensorpack/tfutils/common.py

@@ -46,6 +46,8 @@ def get_default_sess_config(mem_fraction=0.99):
 
     # May hurt performance
     # conf.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1
+    # TODO test this
+    # conf.graph_options.place_pruned_graph = True
     return conf