a rename in examples (not a breaking change)

examples/A3C-Gym/run-atari.py

@@ -38,8 +38,7 @@ def get_player(dumpdir=None):
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         assert NUM_ACTIONS is not None
         return [InputVar(tf.float32, (None,) + IMAGE_SHAPE3, 'state'),
                 InputVar(tf.int32, (None,), 'action'),

examples/A3C-Gym/train-atari.py

@@ -75,8 +75,7 @@ class MySimulatorWorker(SimulatorProcess):
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         assert NUM_ACTIONS is not None
         return [InputVar(tf.float32, (None,) + IMAGE_SHAPE3, 'state'),
                 InputVar(tf.int64, (None,), 'action'),

examples/CTC-TIMIT/train-timit.py

@@ -27,8 +27,7 @@ FEATUREDIM = 39
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, None, FEATUREDIM], 'feat'),   # bxmaxseqx39
                 InputVar(tf.int64, None, 'labelidx'),  # label is b x maxlen, sparse
                 InputVar(tf.int32, None, 'labelvalue'),
@@ -36,8 +35,8 @@ class Model(ModelDesc):
                 InputVar(tf.int32, [None], 'seqlen'),   # b
                 ]
 
-    def _build_graph(self, input_vars):
-        feat, labelidx, labelvalue, labelshape, seqlen = input_vars
+    def _build_graph(self, inputs):
+        feat, labelidx, labelvalue, labelshape, seqlen = inputs
         label = tf.SparseTensor(labelidx, labelvalue, labelshape)
 
         cell = tf.contrib.rnn.BasicLSTMCell(num_units=HIDDEN)

examples/Char-RNN/char-rnn.py

@@ -60,12 +60,12 @@ class CharRNNData(RNGDataFlow):
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.int32, (None, param.seq_len), 'input'),
                 InputVar(tf.int32, (None, param.seq_len), 'nextinput')]
 
-    def _build_graph(self, input_vars):
-        input, nextinput = input_vars
+    def _build_graph(self, inputs):
+        input, nextinput = inputs
 
         cell = tf.contrib.rnn.BasicLSTMCell(num_units=param.rnn_size)
         cell = tf.contrib.rnn.MultiRNNCell([cell] * param.num_rnn_layer)
@@ -131,18 +131,18 @@ def sample(path, start, length):
     ds = CharRNNData(param.corpus, 100000)
 
     model = Model()
-    input_vars = model.get_input_vars()
-    model.build_graph(input_vars, False)
+    inputs = model.get_reuse_placehdrs()
+    model.build_graph(inputs, False)
     sess = tf.Session()
     tfutils.SaverRestore(path).init(sess)
 
     dummy_input = np.zeros((1, 1), dtype='int32')
     with sess.as_default():
         # feed the starting sentence
-        state = model.initial.eval({input_vars[0]: dummy_input})
+        state = model.initial.eval({inputs[0]: dummy_input})
         for c in start[:-1]:
             x = np.array([[ds.lut.get_idx(c)]], dtype='int32')
-            state = model.last_state.eval({input_vars[0]: x, model.initial: state})
+            state = model.last_state.eval({inputs[0]: x, model.initial: state})
 
         def pick(prob):
             t = np.cumsum(prob)
@@ -155,7 +155,7 @@ def sample(path, start, length):
         for k in range(length):
             x = np.array([[ds.lut.get_idx(c)]], dtype='int32')
             [prob, state] = sess.run([model.prob, model.last_state],
-                                     {input_vars[0]: x, model.initial: state})
+                                     {inputs[0]: x, model.initial: state})
             c = ds.lut.get_obj(pick(prob[0]))
             ret += c
         print(ret)

examples/ConvolutionalPoseMachines/load-cpm.py

@@ -44,7 +44,7 @@ def get_gaussian_map():
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, 368, 368, 3), 'input'),
                 InputVar(tf.float32, (None, 368, 368, 15), 'label'),
                 ]

examples/DeepQNetwork/DQN.py

@@ -68,8 +68,7 @@ common.get_player = get_player  # so that eval functions in common can use the p
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         if NUM_ACTIONS is None:
             p = get_player()
             del p

examples/DisturbLabel/mnist-disturb.py

@@ -29,9 +29,8 @@ IMAGE_SIZE = 28
 
 
 class Model(mnist_example.Model):
-
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         image = tf.expand_dims(image, 3)
 
         with argscope(Conv2D, kernel_shape=5, nl=tf.nn.relu):

examples/DoReFa-Net/alexnet-dorefa.py

@@ -74,13 +74,12 @@ BATCH_SIZE = None
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, 224, 224, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         image = image / 255.0
 
         fw, fa, fg = get_dorefa(BITW, BITA, BITG)

examples/DoReFa-Net/resnet-dorefa.py

@@ -33,12 +33,12 @@ BITG = 32
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, 224, 224, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         image = image / 256.0
 
         fw, fa, fg = get_dorefa(BITW, BITA, BITG)

examples/DoReFa-Net/svhn-digit-dorefa.py

@@ -43,13 +43,12 @@ BITG = 4
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, 40, 40, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         is_training = get_current_tower_context().is_training
 
         fw, fa, fg = get_dorefa(BITW, BITA, BITG)

examples/GAN/DCGAN-CelebA.py

@@ -36,8 +36,7 @@ CFG.Z_DIM = 100
 
 
 class Model(GANModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, CFG.SHAPE, CFG.SHAPE, 3), 'input')]
 
     def generator(self, z):
@@ -70,8 +69,8 @@ class Model(GANModelDesc):
                  .FullyConnected('fct', 1, nl=tf.identity)())
         return l
 
-    def _build_graph(self, input_vars):
-        image_pos = input_vars[0]
+    def _build_graph(self, inputs):
+        image_pos = inputs[0]
         image_pos = image_pos / 128.0 - 1
 
         z = tf.random_uniform([CFG.BATCH, CFG.Z_DIM], -1, 1, name='z_train')

examples/GAN/Image2Image.py

@@ -43,8 +43,7 @@ NF = 64  # number of filter
 
 
 class Model(GANModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, SHAPE, SHAPE, IN_CH), 'input'),
                 InputVar(tf.float32, (None, SHAPE, SHAPE, OUT_CH), 'output')]
 
@@ -100,8 +99,8 @@ class Model(GANModelDesc):
                  .Conv2D('convlast', 1, stride=1, padding='VALID')())
         return l
 
-    def _build_graph(self, input_vars):
-        input, output = input_vars
+    def _build_graph(self, inputs):
+        input, output = inputs
         input, output = input / 128.0 - 1, output / 128.0 - 1
 
         with argscope([Conv2D, Deconv2D],

examples/GAN/InfoGAN-mnist.py

@@ -32,7 +32,7 @@ class GaussianWithUniformSample(GaussianDistribution):
 
 
 class Model(GANModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, 28, 28), 'input')]
 
     def generator(self, z):
@@ -62,8 +62,8 @@ class Model(GANModelDesc):
                        .FullyConnected('fce-out', self.factors.param_dim, nl=tf.identity)())
         return logits, encoder
 
-    def _build_graph(self, input_vars):
-        real_sample = input_vars[0]
+    def _build_graph(self, inputs):
+        real_sample = inputs[0]
         real_sample = tf.expand_dims(real_sample * 2.0 - 1, -1)
 
         # latent space is cat(10) x uni(1) x uni(1) x noise(NOISE_DIM)

examples/HED/hed.py

@@ -17,13 +17,12 @@ from tensorpack.tfutils.summary import *
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, None, None, 3], 'image'),
                 InputVar(tf.int32, [None, None, None], 'edgemap')]
 
-    def _build_graph(self, input_vars):
-        image, edgemap = input_vars
+    def _build_graph(self, inputs):
+        image, edgemap = inputs
         image = image - tf.constant([104, 116, 122], dtype='float32')
         edgemap = tf.expand_dims(edgemap, 3, name='edgemap4d')
 

examples/Inception/inception-bn.py

@@ -29,12 +29,12 @@ Learning rate may need a different schedule for different number of GPUs (becaus
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         image = image / 128.0
 
         def inception(name, x, nr1x1, nr3x3r, nr3x3, nr233r, nr233, nrpool, pooltype):

examples/Inception/inceptionv3.py

@@ -35,12 +35,12 @@ INPUT_SHAPE = 299
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         image = image / 255.0   # ?
 
         def proj_kk(l, k, ch_r, ch, stride=1):

examples/PennTreebank/PTB-LSTM.py

@@ -44,13 +44,13 @@ def get_PennTreeBank(data_dir=None):
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.int32, (None, SEQ_LEN), 'input'),
                 InputVar(tf.int32, (None, SEQ_LEN), 'nextinput')]
 
-    def _build_graph(self, input_vars):
+    def _build_graph(self, inputs):
         is_training = get_current_tower_context().is_training
-        input, nextinput = input_vars
+        input, nextinput = inputs
         initializer = tf.random_uniform_initializer(-0.05, 0.05)
 
         cell = rnn.BasicLSTMCell(num_units=HIDDEN_SIZE, forget_bias=0.0)

examples/ResNet/cifar10-resnet.py

@@ -37,12 +37,12 @@ class Model(ModelDesc):
         super(Model, self).__init__()
         self.n = n
 
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, 32, 32, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         image = image / 128.0 - 1
 
         def residual(name, l, increase_dim=False, first=False):

examples/ResNet/imagenet-resnet.py

@@ -28,12 +28,12 @@ DEPTH = None
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
 
         def shortcut(l, n_in, n_out, stride):
             if n_in != n_out:

examples/Saliency/saliency-maps.py

@@ -16,11 +16,11 @@ IMAGE_SIZE = 224
 
 
 class Model(tp.ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [tp.InputVar(tf.float32, (IMAGE_SIZE, IMAGE_SIZE, 3), 'image')]
 
-    def _build_graph(self, input_vars):
-        orig_image = input_vars[0]
+    def _build_graph(self, inputs):
+        orig_image = inputs[0]
         mean = tf.get_variable('resnet_v1_50/mean_rgb', shape=[3])
         with tp.symbolic_functions.guided_relu():
             with slim.arg_scope(resnet_v1.resnet_arg_scope(is_training=False)):

examples/SimilarityLearning/mnist-embeddings.py

@@ -61,20 +61,20 @@ class SiameseModel(EmbeddingModel):
         ds = BatchData(ds, 128 // 2)
         return ds
 
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, 28, 28), 'input'),
                 InputVar(tf.float32, (None, 28, 28), 'input_y'),
                 InputVar(tf.int32, (None,), 'label')]
 
-    def _build_graph(self, input_vars):
+    def _build_graph(self, inputs):
         # get inputs
-        x, y, label = input_vars
+        x, y, label = inputs
         # embed them
         x, y = self.embed([x, y])
 
         # tag the embedding of 'input' with name 'emb', just for inference later on
         with tf.variable_scope(tf.get_variable_scope(), reuse=True):
-            tf.identity(self.embed(input_vars[0]), name="emb")
+            tf.identity(self.embed(inputs[0]), name="emb")
 
         # compute the actual loss
         cost, pos_dist, neg_dist = symbf.contrastive_loss(x, y, label, 5., extra=True)
@@ -85,12 +85,12 @@ class SiameseModel(EmbeddingModel):
 
 
 class CosineModel(SiameseModel):
-    def _build_graph(self, input_vars):
-        x, y, label = input_vars
+    def _build_graph(self, inputs):
+        x, y, label = inputs
         x, y = self.embed([x, y])
 
         with tf.variable_scope(tf.get_variable_scope(), reuse=True):
-            tf.identity(self.embed(input_vars[0]), name="emb")
+            tf.identity(self.embed(inputs[0]), name="emb")
 
         cost = symbf.cosine_loss(x, y, label)
         self.cost = tf.identity(cost, name="cost")
@@ -104,7 +104,7 @@ class TripletModel(EmbeddingModel):
         ds = BatchData(ds, 128 // 3)
         return ds
 
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, 28, 28), 'input'),
                 InputVar(tf.float32, (None, 28, 28), 'input_p'),
                 InputVar(tf.float32, (None, 28, 28), 'input_n')]
@@ -112,12 +112,12 @@ class TripletModel(EmbeddingModel):
     def loss(self, a, p, n):
         return symbf.triplet_loss(a, p, n, 5., extra=True)
 
-    def _build_graph(self, input_vars):
-        a, p, n = input_vars
+    def _build_graph(self, inputs):
+        a, p, n = inputs
         a, p, n = self.embed([a, p, n])
 
         with tf.variable_scope(tf.get_variable_scope(), reuse=True):
-            tf.identity(self.embed(input_vars[0]), name="emb")
+            tf.identity(self.embed(inputs[0]), name="emb")
 
         cost, pos_dist, neg_dist = self.loss(a, p, n)
         self.cost = tf.identity(cost, name="cost")

examples/SpatialTransformer/mnist-addition.py

@@ -18,17 +18,16 @@ HALF_DIFF = (IMAGE_SIZE - WARP_TARGET_SIZE) // 2
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE, 2), 'input'),
                 InputVar(tf.int32, (None,), 'label')]
 
-    def _build_graph(self, input_vars):
+    def _build_graph(self, inputs):
         xys = np.array([(y, x, 1) for y in range(WARP_TARGET_SIZE)
                         for x in range(WARP_TARGET_SIZE)], dtype='float32')
         xys = tf.constant(xys, dtype=tf.float32, name='xys')    # p x 3
 
-        image, label = input_vars
+        image, label = inputs
 
         image = image / 255.0 - 0.5  # bhw2
 

examples/cifar-convnet.py

@@ -24,18 +24,17 @@ Not a good model for Cifar100, just for demonstration.
 
 
 class Model(ModelDesc):
-
     def __init__(self, cifar_classnum):
         super(Model, self).__init__()
         self.cifar_classnum = cifar_classnum
 
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, 30, 30, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')
                 ]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
         is_training = get_current_tower_context().is_training
         keep_prob = tf.constant(0.5 if is_training else 1.0)
 

examples/load-alexnet.py

@@ -24,7 +24,7 @@ Usage:
 
 class Model(ModelDesc):
 
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, 227, 227, 3), 'input')]
 
     def _build_graph(self, inputs):

examples/load-vgg16.py

@@ -24,8 +24,7 @@ Usage:
 
 
 class Model(ModelDesc):
-
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, (None, 224, 224, 3), 'input')]
 
     def _build_graph(self, inputs):

examples/mnist-convnet.py

@@ -23,18 +23,18 @@ USE_SLIM = False
 
 
 class Model(ModelDesc):
-    def _get_input_vars(self):
+    def _get_inputs(self):
         """Define all the input variables (with type, shape, name) that'll be
         fed into the graph to produce a cost.  """
         return [InputVar(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
                 InputVar(tf.int32, (None,), 'label')]
 
-    def _build_graph(self, input_vars):
+    def _build_graph(self, inputs):
         """This function should build the model which takes the input variables
         and define self.cost at the end"""
 
-        # input_vars contains a list of input variables defined above
-        image, label = input_vars
+        # inputs contains a list of input variables defined above
+        image, label = inputs
         # In tensorflow, inputs to convolution function are assumed to be
         # NHWC. Add a single channel here.
         image = tf.expand_dims(image, 3)

examples/svhn-digit-convnet.py

@@ -23,12 +23,12 @@ Speed is about 43 it/s on TitanX.
 
 class Model(ModelDesc):
 
-    def _get_input_vars(self):
+    def _get_inputs(self):
         return [InputVar(tf.float32, [None, 40, 40, 3], 'input'),
                 InputVar(tf.int32, [None], 'label')]
 
-    def _build_graph(self, input_vars):
-        image, label = input_vars
+    def _build_graph(self, inputs):
+        image, label = inputs
 
         image = image / 128.0 - 1
 

tensorpack/models/model_desc.py

@@ -87,11 +87,14 @@ class ModelDesc(object):
         """
         return self._get_input_vars()
 
-    @abstractmethod
-    def _get_input_vars(self):
+    def _get_input_vars(self):  # keep backward compatibility
         """
         :returns: a list of InputVar
         """
+        return self._get_inputs()
+
+    def _get_inputs(self):  # this is a better name than _get_input_vars
+        raise NotImplementedError()
 
     def build_graph(self, model_inputs):
         """
@@ -171,7 +174,7 @@ class ModelFromMetaGraph(ModelDesc):
             assert k in all_coll, \
                 "Collection {} not found in metagraph!".format(k)
 
-    def _get_input_vars(self):
+    def _get_inputs(self):
         col = tf.get_collection(INPUT_VARS_KEY)
         col = [InputVar.loads(v) for v in col]
         return col