update model location

examples/DoReFa-Net/README.md

@@ -5,8 +5,8 @@ Code and model for the paper:
 We hosted a demo at CVPR16 on behalf of Megvii, Inc, running real-time half-VGG size DoReFa-Net on both ARM and FPGA.
 But we're not planning to release those runtime bit-op libraries for now. In these examples, bit operations are run in float32.
 
-Pretrained model for 1-2-6-AlexNet is available
-[here](https://github.com/ppwwyyxx/tensorpack/releases/tag/alexnet-dorefa).
+Pretrained model for 1-2-6-AlexNet is available at
+[google drive](https://drive.google.com/a/%20megvii.com/folderview?id=0B308TeQzmFDLa0xOeVQwcXg1ZjQ).
 It's provided in the format of numpy dictionary, so it should be very easy to port into other applications.
 
 ## Preparation:

examples/ResNet/README.md

@@ -10,4 +10,4 @@ The validation error here is computed on test set.
 ![cifar10](cifar10-resnet.png)
 
 Download model:
-[Cifar10 ResNet-110 (n=18)](https://github.com/ppwwyyxx/tensorpack/releases/tag/cifar10-resnet-110)
+[Cifar10 ResNet-110 (n=18)](https://drive.google.com/open?id=0B9IPQTvr2BBkTXBlZmh1cmlnQ0k)

examples/ResNet/cifar10-resnet.py

@@ -148,6 +148,7 @@ def get_config():
 
     sess_config = get_default_sess_config(0.9)
 
+    get_global_step_var()
     lr = tf.Variable(0.01, trainable=False, name='learning_rate')
     tf.scalar_summary('learning_rate', lr)
 

scripts/dump-model-params.py

@@ -3,6 +3,7 @@
 # File: dump-model-params.py
 # Author: Yuxin Wu <ppwwyyxx@gmail.com>
 
+import numpy as np
 import argparse
 import tensorflow as tf
 import imp
@@ -28,7 +29,10 @@ with tf.Graph().as_default() as G:
         M = ModelFromMetaGraph(args.meta)
 
     # loading...
-    init = sessinit.SaverRestore(args.model)
+    if args.model.endswith('.npy'):
+        init = sessinit.ParamRestore(np.load(args.model).item())
+    else:
+        init = sessinit.SaverRestore(args.model)
     sess = tf.Session()
     init.init(sess)
 

tensorpack/tfutils/common.py

@@ -40,6 +40,9 @@ def get_global_step_var():
     try:
         return tf.get_default_graph().get_tensor_by_name(GLOBAL_STEP_VAR_NAME)
     except KeyError:
+        scope = tf.get_variable_scope()
+        assert scope.name == '', \
+                "Creating global_step_var under a variable scope would cause problems!"
         var = tf.Variable(
             0, trainable=False, name=GLOBAL_STEP_OP_NAME)
         return var