super() in transforms. update docs

CHANGES.md

@@ -6,9 +6,9 @@ The backward compatibilty will be __preserved for at least several months__, wit
 so you don't need to look at here very often.
 
 Here are a list of things that were changed, starting from an early version.
-TensorFlow itself also APIs changes and those are not listed here.
+TensorFlow itself also changes API and those are not listed here.
 
-+ [2018/04/05] msgpack is replaced by pyarrow. If you want compatibility with old serialized data,
++ [2018/04/05] msgpack is replaced by pyarrow. If you want old behavior,
 	`export TENSORPACK_SERIALIZE=msgpack`.
 + [2018/03/20] `ModelDesc` starts to use simplified interfaces:
 	+ `_get_inputs()` renamed to `inputs()` and returns `tf.placeholder`s.

docs/tutorial/inference.md

@@ -23,8 +23,8 @@ You can use this predicate to choose a different code path in inference mode.
 
 Tensorpack is a training interface -- it doesn't care what happened after training.
 It saves models to standard checkpoint format.
-You can build the graph for inference, load the checkpoint, and then use whatever deployment methods TensorFlow supports.
-But you'll need to read TF docs and do it on your own.
+So you can build the graph for inference, load the checkpoint, and then use whatever deployment methods TensorFlow supports.
+But you'll need to read TF docs and __do it on your own__.
 
 ### Don't Use Training Metagraph for Inference
 
@@ -50,6 +50,11 @@ with TowerContext('', is_training=False):
 ### OfflinePredictor
 The only tool tensorpack has for after-training inference is [OfflinePredictor](../modules/predict.html#tensorpack.predict.OfflinePredictor),
 a simple function to build the graph and return a callable for you.
-It is mainly for quick demo purposes.
-It only runs inference on numpy arrays, therefore may not be the most efficient way.
 Check out examples and docs for its usage.
+
+
+OfflinePredictor is only for quick demo purposes.
+It runs inference on numpy arrays, therefore may not be the most efficient way.
+It also has very limited functionalities.
+If you need anything more complicated, please __do it on your own__ because Tensorpack
+doesn't care what happened after training.

examples/FasterRCNN/common.py

@@ -38,6 +38,7 @@ class CustomResize(transform.TransformAugmentorBase):
             size (int): the size to resize the shortest edge to.
             max_size (int): maximum allowed longest edge.
         """
+        super(CustomResize, self).__init__()
         self._init(locals())
 
     def _get_augment_params(self, img):

tensorpack/dataflow/imgaug/transform.py

@@ -56,6 +56,7 @@ class ImageTransform(object):
 
 class ResizeTransform(ImageTransform):
     def __init__(self, h, w, newh, neww, interp):
+        super(ResizeTransform, self).__init__()
         self._init(locals())
 
     def apply_image(self, img):
@@ -75,6 +76,7 @@ class ResizeTransform(ImageTransform):
 
 class CropTransform(ImageTransform):
     def __init__(self, h0, w0, h, w):
+        super(CropTransform, self).__init__()
         self._init(locals())
 
     def apply_image(self, img):
@@ -89,6 +91,7 @@ class CropTransform(ImageTransform):
 class WarpAffineTransform(ImageTransform):
     def __init__(self, mat, dsize, interp=cv2.INTER_LINEAR,
                  borderMode=cv2.BORDER_CONSTANT, borderValue=0):
+        super(WarpAffineTransform, self).__init__()
         self._init(locals())
 
     def apply_image(self, img):

tensorpack/utils/serialize.py

@@ -10,8 +10,9 @@ __all__ = ['loads', 'dumps']
 def dumps_msgpack(obj):
     """
     Serialize an object.
+
     Returns:
-        Implementation-dependent bytes-like object
+        Implementation-dependent bytes-like object.
     """
     return msgpack.dumps(obj, use_bin_type=True)
 
@@ -29,7 +30,8 @@ def dumps_pyarrow(obj):
     Serialize an object.
 
     Returns:
-        Implementation-dependent bytes-like object
+        Implementation-dependent bytes-like object.
+        May not be compatible across different versions of pyarrow.
     """
     return pa.serialize(obj).to_buffer()