1.0.0rc0 change concat_v2 to concat (#121)

README.md

@@ -58,7 +58,7 @@ The components are designed to be independent. You can use Model or DataFlow in
 ## Dependencies:
 
 + Python 2 or 3
-+ TensorFlow >= 0.12
++ TensorFlow >= 1.0.0rc0
 + Python bindings for OpenCV
 + other requirements:
 ```

examples/Char-RNN/char-rnn.py

@@ -84,7 +84,7 @@ class Model(ModelDesc):
         self.last_state = tf.identity(last_state, 'last_state')
 
         # seqlen x (Bxrnnsize)
-        output = tf.reshape(tf.concat_v2(outputs, 1), [-1, param.rnn_size])  # (Bxseqlen) x rnnsize
+        output = tf.reshape(tf.concat(outputs, 1), [-1, param.rnn_size])  # (Bxseqlen) x rnnsize
         logits = FullyConnected('fc', output, param.vocab_size, nl=tf.identity)
         self.prob = tf.nn.softmax(logits / param.softmax_temprature)
 

examples/ConvolutionalPoseMachines/load-cpm.py

@@ -82,7 +82,7 @@ class Model(ModelDesc):
                       .Conv2D('conv4_7_CPM', 128)())
 
         def add_stage(stage, l):
-            l = tf.concat_v2([l, shared, pool_center], 3,
+            l = tf.concat([l, shared, pool_center], 3,
                              name='concat_stage{}'.format(stage))
             for i in range(1, 6):
                 l = Conv2D('Mconv{}_stage{}'.format(i, stage), l, 128)

examples/DoReFa-Net/README.md

@@ -20,7 +20,7 @@ Alternative link to this page: [http://dorefa.net](http://dorefa.net)
 
 To use the script. You'll need:
 
-+ TensorFlow >= 0.12.1
++ TensorFlow >= 1.0.0rc0
 
 + OpenCV bindings for Python
 

examples/GAN/Image2Image.py

@@ -87,7 +87,7 @@ class Model(GANModelDesc):
 
     def discriminator(self, inputs, outputs):
         """ return a (b, 1) logits"""
-        l = tf.concat_v2([inputs, outputs], 3)
+        l = tf.concat([inputs, outputs], 3)
         with argscope(Conv2D, nl=tf.identity, kernel_shape=4, stride=2):
             l = (LinearWrap(l)
                  .Conv2D('conv0', NF, nl=LeakyReLU)
@@ -125,7 +125,7 @@ class Model(GANModelDesc):
         if OUT_CH == 1:
             output = tf.image.grayscale_to_rgb(output)
             fake_output = tf.image.grayscale_to_rgb(fake_output)
-        viz = (tf.concat_v2([input, output, fake_output], 2) + 1.0) * 128.0
+        viz = (tf.concat([input, output, fake_output], 2) + 1.0) * 128.0
         viz = tf.cast(tf.clip_by_value(viz, 0, 255), tf.uint8, name='viz')
         tf.summary.image('input,output,fake', viz, max_outputs=max(30, BATCH))
 

examples/GAN/InfoGAN-mnist.py

@@ -79,7 +79,7 @@ class Model(GANModelDesc):
             self.factors.sample(BATCH, prior), 0, name='z_code')
         z_noise = symbf.shapeless_placeholder(
             tf.random_uniform([BATCH, NOISE_DIM], -1, 1), 0, name='z_noise')
-        z = tf.concat_v2([zc, z_noise], 1, name='z')
+        z = tf.concat([zc, z_noise], 1, name='z')
 
         with argscope([Conv2D, Deconv2D, FullyConnected],
                       W_init=tf.truncated_normal_initializer(stddev=0.02)):

examples/HED/hed.py

@@ -67,7 +67,7 @@ class Model(ModelDesc):
             b5 = branch('branch5', l, 16)
 
         final_map = Conv2D('convfcweight',
-                           tf.concat_v2([b1, b2, b3, b4, b5], 3), 1, 1,
+                           tf.concat([b1, b2, b3, b4, b5], 3), 1, 1,
                            W_init=tf.constant_initializer(0.2),
                            use_bias=False, nl=tf.identity)
         costs = []

examples/Inception/inception-bn.py

@@ -59,7 +59,7 @@ class Model(ModelDesc):
                 if nrpool != 0:  # pool + passthrough if nrpool == 0
                     x4 = Conv2D('poolproj', x4, nrpool, 1)
                 outs.append(x4)
-                return tf.concat_v2(outs, 3, name='concat')
+                return tf.concat(outs, 3, name='concat')
 
         with argscope(Conv2D, nl=BNReLU, use_bias=False):
             l = Conv2D('conv0', image, 64, 7, stride=2)

examples/Inception/inceptionv3.py

@@ -88,21 +88,21 @@ class Model(ModelDesc):
                  .MaxPooling('pool4', 3, 2)())  # 35
 
             with tf.variable_scope('incep-35-256a'):
-                l = tf.concat_v2([
+                l = tf.concat([
                     Conv2D('conv11', l, 64, 1),
                     proj_kk(l, 5, 48, 64),
                     proj_233(l, 64, 96),
                     pool_proj(l, 32, 'avg')
                 ], 3, name='concat')
             with tf.variable_scope('incep-35-288a'):
-                l = tf.concat_v2([
+                l = tf.concat([
                     Conv2D('conv11', l, 64, 1),
                     proj_kk(l, 5, 48, 64),
                     proj_233(l, 64, 96),
                     pool_proj(l, 64, 'avg')
                 ], 3, name='concat')
             with tf.variable_scope('incep-35-288b'):
-                l = tf.concat_v2([
+                l = tf.concat([
                     Conv2D('conv11', l, 64, 1),
                     proj_kk(l, 5, 48, 64),
                     proj_233(l, 64, 96),
@@ -110,13 +110,13 @@ class Model(ModelDesc):
                 ], 3, name='concat')
             # 35x35x288
             with tf.variable_scope('incep-17-768a'):
-                l = tf.concat_v2([
+                l = tf.concat([
                     Conv2D('conv3x3', l, 384, 3, stride=2, padding='VALID'),
                     proj_233(l, 64, 96, stride=2),
                     MaxPooling('maxpool', l, 3, 2)
                 ], 3, name='concat')
             with tf.variable_scope('incep-17-768b'):
-                l = tf.concat_v2([
+                l = tf.concat([
                     Conv2D('conv11', l, 192, 1),
                     proj_77(l, 128, 192),
                     proj_277(l, 128, 192),
@@ -124,14 +124,14 @@ class Model(ModelDesc):
                 ], 3, name='concat')
             for x in ['c', 'd']:
                 with tf.variable_scope('incep-17-768{}'.format(x)):
-                    l = tf.concat_v2([
+                    l = tf.concat([
                         Conv2D('conv11', l, 192, 1),
                         proj_77(l, 160, 192),
                         proj_277(l, 160, 192),
                         pool_proj(l, 192, 'avg')
                     ], 3, name='concat')
             with tf.variable_scope('incep-17-768e'):
-                l = tf.concat_v2([
+                l = tf.concat([
                     Conv2D('conv11', l, 192, 1),
                     proj_77(l, 192, 192),
                     proj_277(l, 192, 192),
@@ -147,7 +147,7 @@ class Model(ModelDesc):
                 br1 = FullyConnected('fc', br1, 1000, nl=tf.identity)
 
             with tf.variable_scope('incep-17-1280a'):
-                l = tf.concat_v2([
+                l = tf.concat([
                     proj_kk(l, 3, 192, 320, stride=2),
                     Conv2D('conv73', proj_77(l, 192, 192), 192, 3, stride=2, padding='VALID'),
                     MaxPooling('maxpool', l, 3, 2)
@@ -156,18 +156,18 @@ class Model(ModelDesc):
                 with tf.variable_scope('incep-8-2048{}'.format(x)) as scope:
                     br11 = Conv2D('conv11', l, 320, 1)
                     br33 = Conv2D('conv133r', l, 384, 1)
-                    br33 = tf.concat_v2([
+                    br33 = tf.concat([
                         Conv2D('conv133a', br33, 384, [1, 3]),
                         Conv2D('conv133b', br33, 384, [3, 1])
                     ], 3, name='conv133')
 
                     br233 = proj_kk(l, 3, 448, 384)
-                    br233 = tf.concat_v2([
+                    br233 = tf.concat([
                         Conv2D('conv233a', br233, 384, [1, 3]),
                         Conv2D('conv233b', br233, 384, [3, 1]),
                     ], 3, name='conv233')
 
-                    l = tf.concat_v2([
+                    l = tf.concat([
                         br11, br33, br233,
                         pool_proj(l, 192, 'avg')
                     ], 3, name='concat')

examples/PennTreebank/PTB-LSTM.py

@@ -75,7 +75,7 @@ class Model(ModelDesc):
             outputs, last_state = rnn.static_rnn(cell, input_list, state_var, scope='rnn')
 
         # seqlen x (Bxrnnsize)
-        output = tf.reshape(tf.concat_v2(outputs, 1), [-1, HIDDEN_SIZE])  # (Bxseqlen) x hidden
+        output = tf.reshape(tf.concat(outputs, 1), [-1, HIDDEN_SIZE])  # (Bxseqlen) x hidden
         logits = FullyConnected('fc', output, VOCAB_SIZE, nl=tf.identity, W_init=initializer, b_init=initializer)
         xent_loss = tf.nn.sparse_softmax_cross_entropy_with_logits(
             logits=logits, labels=symbolic_functions.flatten(nextinput))

examples/SimilarityLearning/mnist-embeddings.py

@@ -32,7 +32,7 @@ class EmbeddingModel(ModelDesc):
         list_split = 0
         if isinstance(x, list):
             list_split = len(x)
-            x = tf.concat_v2(x, 0)
+            x = tf.concat(x, 0)
 
         # pre-process MNIST dataflow data
         x = tf.expand_dims(x, 3)

examples/SpatialTransformer/mnist-addition.py

@@ -60,14 +60,14 @@ class Model(ModelDesc):
         # For visualization in tensorboard
         padded1 = tf.pad(sampled1, [[0, 0], [HALF_DIFF, HALF_DIFF], [HALF_DIFF, HALF_DIFF], [0, 0]])
         padded2 = tf.pad(sampled2, [[0, 0], [HALF_DIFF, HALF_DIFF], [HALF_DIFF, HALF_DIFF], [0, 0]])
-        img_orig = tf.concat_v2([image[:, :, :, 0], image[:, :, :, 1]], 1)  # b x 2h  x w
-        transform1 = tf.concat_v2([padded1[:, :, :, 0], padded1[:, :, :, 1]], 1)
-        transform2 = tf.concat_v2([padded2[:, :, :, 0], padded2[:, :, :, 1]], 1)
-        stacked = tf.concat_v2([img_orig, transform1, transform2], 2, 'viz')
+        img_orig = tf.concat([image[:, :, :, 0], image[:, :, :, 1]], 1)  # b x 2h  x w
+        transform1 = tf.concat([padded1[:, :, :, 0], padded1[:, :, :, 1]], 1)
+        transform2 = tf.concat([padded2[:, :, :, 0], padded2[:, :, :, 1]], 1)
+        stacked = tf.concat([img_orig, transform1, transform2], 2, 'viz')
         tf.summary.image('visualize',
                          tf.expand_dims(stacked, -1), max_outputs=30)
 
-        sampled = tf.concat_v2([sampled1, sampled2], 3, 'sampled_concat')
+        sampled = tf.concat([sampled1, sampled2], 3, 'sampled_concat')
         logits = (LinearWrap(sampled)
                   .apply(symbf.batch_flatten)
                   .FullyConnected('fc1', out_dim=256, nl=tf.nn.relu)

tensorpack/models/conv2d.py

@@ -66,7 +66,7 @@ def Conv2D(x, out_channel, kernel_shape,
         kernels = tf.split(W, split, 3)
         outputs = [tf.nn.conv2d(i, k, stride, padding)
                    for i, k in zip(inputs, kernels)]
-        conv = tf.concat_v2(outputs, 3)
+        conv = tf.concat(outputs, 3)
     return nl(tf.nn.bias_add(conv, b) if use_bias else conv, name='output')
 
 

tensorpack/models/image_sample.py

@@ -80,8 +80,8 @@ def ImageSample(inputs, borderMode='repeat'):
     lcoory, lcoorx = tf.split(lcoor, 2, 3)
     ucoory, ucoorx = tf.split(ucoor, 2, 3)
 
-    lyux = tf.concat_v2([lcoory, ucoorx], 3)
-    uylx = tf.concat_v2([ucoory, lcoorx], 3)
+    lyux = tf.concat([lcoory, ucoorx], 3)
+    uylx = tf.concat([ucoory, lcoorx], 3)
 
     diffy, diffx = tf.split(diff, 2, 3)
     neg_diffy, neg_diffx = tf.split(neg_diff, 2, 3)

tensorpack/models/pool.py

@@ -83,8 +83,8 @@ def GlobalAvgPooling(x):
 
 def UnPooling2x2ZeroFilled(x):
     # https://github.com/tensorflow/tensorflow/issues/2169
-    out = tf.concat_v2([x, tf.zeros_like(x)], 3)
-    out = tf.concat_v2([out, tf.zeros_like(out)], 2)
+    out = tf.concat([x, tf.zeros_like(x)], 3)
+    out = tf.concat([out, tf.zeros_like(out)], 2)
 
     sh = x.get_shape().as_list()
     if None not in sh[1:]:

tensorpack/models/shapes.py

@@ -12,7 +12,7 @@ __all__ = ['ConcatWith']
 @layer_register(use_scope=False, log_shape=False)
 def ConcatWith(x, tensor, dim):
     """
-    A wrapper around ``tf.concat_v2`` to cooperate with :class:`LinearWrap`.
+    A wrapper around ``tf.concat`` to cooperate with :class:`LinearWrap`.
 
     Args:
         x (tf.Tensor): input
@@ -21,8 +21,8 @@ def ConcatWith(x, tensor, dim):
         dim (int): the dimension along which to concatenate
 
     Returns:
-        tf.Tensor: ``tf.concat_v2([x] + tensor, dim)``
+        tf.Tensor: ``tf.concat([x] + tensor, dim)``
     """
     if type(tensor) != list:
         tensor = [tensor]
-    return tf.concat_v2([x] + tensor, dim)
+    return tf.concat([x] + tensor, dim)

tensorpack/tfutils/distributions.py

@@ -225,7 +225,7 @@ class GaussianDistribution(Distribution):
             stddev = tf.exp(stddev)  # just make it positive and assume it's stddev
             # OpenAI code assumes exp(input) is variance. https://github.com/openai/InfoGAN.
             # not sure if there is any theory about this.
-            return tf.concat_v2([mean, stddev], axis=1)
+            return tf.concat([mean, stddev], axis=1)
 
     def _sample(self, batch_size, theta):
         if self.fixed_std:
@@ -308,10 +308,10 @@ class ProductDistribution(Distribution):
         for dist, dist_param in zip(self.dists, self._splitter(dist_params, True)):
             if dist.param_dim > 0:
                 rsl.append(dist._encoder_activation(dist_param))
-        return tf.concat_v2(rsl, 1)
+        return tf.concat(rsl, 1)
 
     def _sample(self, batch_size, theta):
         ret = []
         for dist, ti in zip(self.dists, self._splitter(theta, True)):
             ret.append(dist._sample(batch_size, ti))
-        return tf.concat_v2(ret, 1, name='sample')
+        return tf.concat(ret, 1, name='sample')

tensorpack/tfutils/symbolic_functions.py

@@ -130,7 +130,7 @@ def huber_loss(x, delta=1, name='huber_loss'):
 
 def get_scalar_var(name, init_value, summary=False, trainable=False):
     """
-    Get a scalar variable with certain initial value
+    Get a scalar float variable with certain initial value
 
     Args:
         name (str): name of the variable.
@@ -140,7 +140,7 @@ def get_scalar_var(name, init_value, summary=False, trainable=False):
     Returns:
         tf.Variable: the variable
     """
-    ret = tf.get_variable(name, initializer=init_value,
+    ret = tf.get_variable(name, initializer=float(init_value),
                           trainable=trainable)
     if summary:
         # this is recognized in callbacks.StatHolder