update docs

README.md

@@ -68,7 +68,7 @@ Dependencies:
 
 + Python 2.7 or 3.3+. Python 2.7 is supported until [it retires in 2020](https://pythonclock.org/).
 + Python bindings for OpenCV. (Optional, but required by a lot of features)
-+ TensorFlow ≥ 1.3, < 2. (Optional, if you only want to use `tensorpack.dataflow` alone as a data processing library)
++ TensorFlow ≥ 1.3, < 2. (Not required if you only want to use `tensorpack.dataflow` alone as a data processing library)
 ```
 pip install --upgrade git+https://github.com/tensorpack/tensorpack.git
 # or add `--user` to install to user's local directories

docs/tutorial/inference.md

@@ -40,15 +40,15 @@ for inference demo after training.
 It has functionailities to build the graph, load the checkpoint, and
 return a callable for you for simple prediction. Refer to its docs for details.
 
-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.
+To use it, you need to provide your model, checkpoint, and define what are the
+input & output tensors to infer with. You can obtain names of tensors by
+`print()`, or assign a name to a tensor by `tf.identity(x, name=)`.
 
 A simple example of how it works:
 ```python
 pred_config = PredictConfig(
-    session_init=get_model_loader(model_path),
     model=YourModel(),
+    session_init=get_model_loader(model_path),
     input_names=['input1', 'input2'],  # tensor names in the graph, or name of the declared inputs
     output_names=['output1', 'output2'])  # tensor names in the graph
 predictor = OfflinePredictor(pred_config)
@@ -60,6 +60,10 @@ e.g., use NHWC format, support encoded image format, etc.
 You can make these changes inside the `model` or `tower_func` in your `PredictConfig`.
 The example in [examples/basics/export-model.py](../examples/basics/export-model.py) demonstrates such an altered inference graph.
 
+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.
+
 ### Exporter
 
 In addition to the standard checkpoint format tensorpack saved for you during training,

tensorpack/models/conv2d.py

@@ -37,12 +37,11 @@ def Conv2D(
         activity_regularizer=None,
         split=1):
     """
-    A wrapper around `tf.layers.Conv2D`.
-    Some differences to maintain backward-compatibility:
+    Similar to `tf.layers.Conv2D`, but with some differences:
 
     1. Default kernel initializer is variance_scaling_initializer(2.0).
     2. Default padding is 'same'.
-    3. Support 'split' argument to do group conv. Note that this is not efficient.
+    3. Support 'split' argument to do group convolution.
 
     Variable Names:
 

tensorpack/models/nonlin.py

@@ -4,6 +4,7 @@
 
 import tensorflow as tf
 
+from ..utils.develop import log_deprecated
 from ..compat import tfv1
 from .batch_norm import BatchNorm
 from .common import VariableHolder, layer_register
@@ -36,7 +37,7 @@ def Maxout(x, num_unit):
 
 
 @layer_register()
-def PReLU(x, init=0.001, name='output'):
+def PReLU(x, init=0.001, name=None):
     """
     Parameterized ReLU as in the paper `Delving Deep into Rectifiers: Surpassing
     Human-Level Performance on ImageNet Classification
@@ -45,16 +46,18 @@ def PReLU(x, init=0.001, name='output'):
     Args:
         x (tf.Tensor): input
         init (float): initial value for the learnable slope.
-        name (str): name of the output.
+        name (str): deprecated argument. Don't use
 
     Variable Names:
 
     * ``alpha``: learnable slope.
     """
+    if name is not None:
+        log_deprecated("PReLU(name=...) is deprecated! The output tensor will be named `output`.")
     init = tfv1.constant_initializer(init)
     alpha = tfv1.get_variable('alpha', [], initializer=init)
     x = ((1 + alpha) * x + (1 - alpha) * tf.abs(x))
-    ret = tf.multiply(x, 0.5, name=name)
+    ret = tf.multiply(x, 0.5, name=name or None)
 
     ret.variables = VariableHolder(alpha=alpha)
     return ret
@@ -64,7 +67,14 @@ def PReLU(x, init=0.001, name='output'):
 def BNReLU(x, name=None):
     """
     A shorthand of BatchNormalization + ReLU.
+
+    Args:
+        x (tf.Tensor): the input
+        name: deprecated, don't use.
     """
+    if name is not None:
+        log_deprecated("BNReLU(name=...) is deprecated! The output tensor will be named `output`.")
+
     x = BatchNorm('bn', x)
     x = tf.nn.relu(x, name=name)
     return x