sed -i 's/InputDesc/tf.placeholder/g;s/_get_inputs/inputs/g' (#318)

examples/A3C-Gym/train-atari.py

@@ -69,12 +69,12 @@ class MySimulatorWorker(SimulatorProcess):
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
+    def inputs(self):
         assert NUM_ACTIONS is not None
-        return [InputDesc(tf.uint8, (None,) + IMAGE_SHAPE3, 'state'),
-                InputDesc(tf.int64, (None,), 'action'),
-                InputDesc(tf.float32, (None,), 'futurereward'),
-                InputDesc(tf.float32, (None,), 'action_prob'),
+        return [tf.placeholder(tf.uint8, (None,) + IMAGE_SHAPE3, 'state'),
+                tf.placeholder(tf.int64, (None,), 'action'),
+                tf.placeholder(tf.float32, (None,), 'futurereward'),
+                tf.placeholder(tf.float32, (None,), 'action_prob'),
                 ]
 
     def _get_NN_prediction(self, image):

examples/CTC-TIMIT/train-timit.py

@@ -25,12 +25,12 @@ FEATUREDIM = 39     # MFCC feature dimension
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, None, FEATUREDIM], 'feat'),   # bxmaxseqx39
-                InputDesc(tf.int64, [None, None], 'labelidx'),  # label is b x maxlen, sparse
-                InputDesc(tf.int32, [None], 'labelvalue'),
-                InputDesc(tf.int64, [None], 'labelshape'),
-                InputDesc(tf.int32, [None], 'seqlen'),   # b
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, None, FEATUREDIM], 'feat'),   # bxmaxseqx39
+                tf.placeholder(tf.int64, [None, None], 'labelidx'),  # label is b x maxlen, sparse
+                tf.placeholder(tf.int32, [None], 'labelvalue'),
+                tf.placeholder(tf.int64, [None], 'labelshape'),
+                tf.placeholder(tf.int32, [None], 'seqlen'),   # b
                 ]
 
     def _build_graph(self, inputs):

examples/Char-RNN/char-rnn.py

@@ -70,9 +70,9 @@ class CharRNNData(RNGDataFlow):
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.int32, (None, param.seq_len), 'input'),
-                InputDesc(tf.int32, (None, param.seq_len), 'nextinput')]
+    def inputs(self):
+        return [tf.placeholder(tf.int32, (None, param.seq_len), 'input'),
+                tf.placeholder(tf.int32, (None, param.seq_len), 'nextinput')]
 
     def _build_graph(self, inputs):
         input, nextinput = inputs

examples/DeepQNetwork/DQNModel.py

@@ -6,7 +6,7 @@
 import abc
 import tensorflow as tf
 import tensorpack
-from tensorpack import ModelDesc, InputDesc
+from tensorpack import ModelDesc
 from tensorpack.utils import logger
 from tensorpack.tfutils import (
     varreplace, summary, get_current_tower_context, optimizer, gradproc)
@@ -24,15 +24,15 @@ class Model(ModelDesc):
         self.num_actions = num_actions
         self.gamma = gamma
 
-    def _get_inputs(self):
+    def inputs(self):
         # Use a combined state for efficiency.
         # The first h channels are the current state, and the last h channels are the next state.
-        return [InputDesc(tf.uint8,
-                          (None,) + self.image_shape + (self.channel + 1,),
-                          'comb_state'),
-                InputDesc(tf.int64, (None,), 'action'),
-                InputDesc(tf.float32, (None,), 'reward'),
-                InputDesc(tf.bool, (None,), 'isOver')]
+        return [tf.placeholder(tf.uint8,
+                               (None,) + self.image_shape + (self.channel + 1,),
+                               'comb_state'),
+                tf.placeholder(tf.int64, (None,), 'action'),
+                tf.placeholder(tf.float32, (None,), 'reward'),
+                tf.placeholder(tf.bool, (None,), 'isOver')]
 
     @abc.abstractmethod
     def _get_DQN_prediction(self, image):

examples/DoReFa-Net/alexnet-dorefa.py

@@ -77,9 +77,9 @@ BATCH_SIZE = None
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, 224, 224, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, 224, 224, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/DoReFa-Net/resnet-dorefa.py

@@ -32,9 +32,9 @@ BITG = 32
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, 224, 224, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, 224, 224, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

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

@@ -43,9 +43,9 @@ BITG = 4
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, 40, 40, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, 40, 40, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/DynamicFilterNetwork/steering-filter.py

@@ -95,11 +95,11 @@ class OnlineTensorboardExport(Callback):
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (BATCH, ), 'theta'),
-                InputDesc(tf.float32, (BATCH, SHAPE, SHAPE), 'image'),
-                InputDesc(tf.float32, (BATCH, SHAPE, SHAPE), 'gt_image'),
-                InputDesc(tf.float32, (BATCH, 9, 9), 'gt_filter')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (BATCH, ), 'theta'),
+                tf.placeholder(tf.float32, (BATCH, SHAPE, SHAPE), 'image'),
+                tf.placeholder(tf.float32, (BATCH, SHAPE, SHAPE), 'gt_image'),
+                tf.placeholder(tf.float32, (BATCH, 9, 9), 'gt_filter')]
 
     def _parameter_net(self, theta, kernel_shape=9):
         """Estimate filters for convolution layers

examples/FasterRCNN/train.py

@@ -57,16 +57,16 @@ def get_model_output_names():
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
+    def inputs(self):
         ret = [
-            InputDesc(tf.float32, (None, None, 3), 'image'),
-            InputDesc(tf.int32, (None, None, config.NUM_ANCHOR), 'anchor_labels'),
-            InputDesc(tf.float32, (None, None, config.NUM_ANCHOR, 4), 'anchor_boxes'),
-            InputDesc(tf.float32, (None, 4), 'gt_boxes'),
-            InputDesc(tf.int64, (None,), 'gt_labels')]  # all > 0
+            tf.placeholder(tf.float32, (None, None, 3), 'image'),
+            tf.placeholder(tf.int32, (None, None, config.NUM_ANCHOR), 'anchor_labels'),
+            tf.placeholder(tf.float32, (None, None, config.NUM_ANCHOR, 4), 'anchor_boxes'),
+            tf.placeholder(tf.float32, (None, 4), 'gt_boxes'),
+            tf.placeholder(tf.int64, (None,), 'gt_labels')]  # all > 0
         if config.MODE_MASK:
             ret.append(
-                InputDesc(tf.uint8, (None, None, None), 'gt_masks')
+                tf.placeholder(tf.uint8, (None, None, None), 'gt_masks')
             )   # NR_GT x height x width
         return ret
 

examples/GAN/BEGAN.py

@@ -26,8 +26,8 @@ GAMMA = 0.5
 
 
 class Model(GANModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, args.final_size, args.final_size, 3), 'input')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, args.final_size, args.final_size, 3), 'input')]
 
     @auto_reuse_variable_scope
     def decoder(self, z):

examples/GAN/ConditionalGAN-mnist.py

@@ -40,9 +40,9 @@ def batch_flatten(x):
 
 
 class Model(GANModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, 28, 28), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, 28, 28), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def generator(self, z, y):
         l = FullyConnected('fc0', tf.concat([z, y], 1), 1024, activation=BNReLU)

examples/GAN/CycleGAN.py

@@ -41,9 +41,9 @@ def INLReLU(x, name=None):
 
 
 class Model(GANModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, SHAPE, SHAPE, 3), 'inputA'),
-                InputDesc(tf.float32, (None, SHAPE, SHAPE, 3), 'inputB')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, SHAPE, SHAPE, 3), 'inputA'),
+                tf.placeholder(tf.float32, (None, SHAPE, SHAPE, 3), 'inputB')]
 
     @staticmethod
     def build_res_block(x, name, chan, first=False):

examples/GAN/DCGAN.py

@@ -40,8 +40,8 @@ class Model(GANModelDesc):
         self.batch = batch
         self.zdim = z_dim
 
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, self.shape, self.shape, 3), 'input')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, self.shape, self.shape, 3), 'input')]
 
     def generator(self, z):
         """ return an image generated from z"""

examples/GAN/DiscoGAN-CelebA.py

@@ -34,9 +34,9 @@ def BNLReLU(x, name=None):
 
 
 class Model(GANModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, SHAPE, SHAPE, 3), 'inputA'),
-                InputDesc(tf.float32, (None, SHAPE, SHAPE, 3), 'inputB')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, SHAPE, SHAPE, 3), 'inputA'),
+                tf.placeholder(tf.float32, (None, SHAPE, SHAPE, 3), 'inputB')]
 
     @auto_reuse_variable_scope
     def generator(self, img):

examples/GAN/Image2Image.py

@@ -63,10 +63,10 @@ def visualize_tensors(name, imgs, scale_func=lambda x: (x + 1.) * 128., max_outp
 
 
 class Model(GANModelDesc):
-    def _get_inputs(self):
+    def inputs(self):
         SHAPE = 256
-        return [InputDesc(tf.float32, (None, SHAPE, SHAPE, IN_CH), 'input'),
-                InputDesc(tf.float32, (None, SHAPE, SHAPE, OUT_CH), 'output')]
+        return [tf.placeholder(tf.float32, (None, SHAPE, SHAPE, IN_CH), 'input'),
+                tf.placeholder(tf.float32, (None, SHAPE, SHAPE, OUT_CH), 'output')]
 
     def generator(self, imgs):
         # imgs: input: 256x256xch

examples/GAN/InfoGAN-mnist.py

@@ -105,8 +105,8 @@ def sample_prior(batch_size):
 
 
 class Model(GANModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, 28, 28), 'input')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, 28, 28), 'input')]
 
     def generator(self, z):
         l = FullyConnected('fc0', z, 1024, activation=BNReLU)

examples/HED/hed.py

@@ -44,9 +44,9 @@ def class_balanced_sigmoid_cross_entropy(logits, label, name='cross_entropy_loss
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, None, None, 3], 'image'),
-                InputDesc(tf.int32, [None, None, None], 'edgemap')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, None, None, 3], 'image'),
+                tf.placeholder(tf.int32, [None, None, None], 'edgemap')]
 
     def _build_graph(self, inputs):
         image, edgemap = inputs

examples/ImageNetModels/imagenet_utils.py

@@ -9,7 +9,7 @@ import multiprocessing
 import tensorflow as tf
 from abc import abstractmethod
 
-from tensorpack import imgaug, dataset, ModelDesc, InputDesc
+from tensorpack import imgaug, dataset, ModelDesc
 from tensorpack.dataflow import (
     AugmentImageComponent, PrefetchDataZMQ,
     BatchData, MultiThreadMapData)
@@ -148,9 +148,9 @@ class ImageNetModel(ModelDesc):
     def __init__(self, data_format='NCHW'):
         self.data_format = data_format
 
-    def _get_inputs(self):
-        return [InputDesc(self.image_dtype, [None, self.image_shape, self.image_shape, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(self.image_dtype, [None, self.image_shape, self.image_shape, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/ImageNetModels/inception-bn.py

@@ -24,9 +24,9 @@ INPUT_SHAPE = 224
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/PennTreebank/PTB-LSTM.py

@@ -46,9 +46,9 @@ def get_PennTreeBank(data_dir=None):
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.int32, (None, SEQ_LEN), 'input'),
-                InputDesc(tf.int32, (None, SEQ_LEN), 'nextinput')]
+    def inputs(self):
+        return [tf.placeholder(tf.int32, (None, SEQ_LEN), 'input'),
+                tf.placeholder(tf.int32, (None, SEQ_LEN), 'nextinput')]
 
     def _build_graph(self, inputs):
         is_training = get_current_tower_context().is_training

examples/ResNet/cifar10-preact18-mixup.py

@@ -39,9 +39,9 @@ def preactivation_block(input, num_filters, stride=1):
 
 
 class ResNet_Cifar(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, 32, 32, 3], 'input'),
-                InputDesc(tf.float32, [None, CLASS_NUM], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, 32, 32, 3], 'input'),
+                tf.placeholder(tf.float32, [None, CLASS_NUM], 'label')]
 
     def _build_graph(self, inputs):
         assert tf.test.is_gpu_available()

examples/ResNet/cifar10-resnet.py

@@ -40,9 +40,9 @@ class Model(ModelDesc):
         super(Model, self).__init__()
         self.n = n
 
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, 32, 32, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, 32, 32, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/ResNet/load-resnet.py

@@ -28,9 +28,9 @@ CFG = {
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, 224, 224, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, 224, 224, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/Saliency/CAM-resnet.py

@@ -31,9 +31,9 @@ DEPTH = None
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.uint8, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.uint8, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/Saliency/saliency-maps.py

@@ -54,8 +54,8 @@ def saliency_map(output, input, name="saliency_map"):
 
 
 class Model(tp.ModelDesc):
-    def _get_inputs(self):
-        return [tp.InputDesc(tf.float32, (IMAGE_SIZE, IMAGE_SIZE, 3), 'image')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (IMAGE_SIZE, IMAGE_SIZE, 3), 'image')]
 
     def _build_graph(self, inputs):
         orig_image = inputs[0]

examples/SimilarityLearning/mnist-embeddings.py

@@ -236,10 +236,10 @@ class SiameseModel(EmbeddingModel):
         ds = BatchData(ds, 128 // 2)
         return ds
 
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, 28, 28), 'input'),
-                InputDesc(tf.float32, (None, 28, 28), 'input_y'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, 28, 28), 'input'),
+                tf.placeholder(tf.float32, (None, 28, 28), 'input_y'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
         # get inputs
@@ -279,10 +279,10 @@ class TripletModel(EmbeddingModel):
         ds = BatchData(ds, 128 // 3)
         return ds
 
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, 28, 28), 'input'),
-                InputDesc(tf.float32, (None, 28, 28), 'input_p'),
-                InputDesc(tf.float32, (None, 28, 28), 'input_n')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, 28, 28), 'input'),
+                tf.placeholder(tf.float32, (None, 28, 28), 'input_p'),
+                tf.placeholder(tf.float32, (None, 28, 28), 'input_n')]
 
     def loss(self, a, p, n):
         return triplet_loss(a, p, n, 5., extra=True, scope="loss")
@@ -312,9 +312,9 @@ class CenterModel(EmbeddingModel):
         ds = BatchData(ds, 128)
         return ds
 
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, 28, 28), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, 28, 28), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
         # get inputs

examples/SpatialTransformer/mnist-addition.py

@@ -20,9 +20,9 @@ HALF_DIFF = (IMAGE_SIZE - WARP_TARGET_SIZE) // 2
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE, 2), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE, 2), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
         xys = np.array([(y, x, 1) for y in range(WARP_TARGET_SIZE)

examples/SuperResolution/enet-pat.py

@@ -48,9 +48,9 @@ class Model(GANModelDesc):
         self.height = height
         self.width = width
 
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, self.height * 1, self.width * 1, CHANNELS), 'Ilr'),
-                InputDesc(tf.float32, (None, self.height * 4, self.width * 4, CHANNELS), 'Ihr')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, self.height * 1, self.width * 1, CHANNELS), 'Ilr'),
+                tf.placeholder(tf.float32, (None, self.height * 4, self.width * 4, CHANNELS), 'Ihr')]
 
     def _build_graph(self, inputs):
         ctx = get_current_tower_context()

examples/basics/mnist-tflayers.py

@@ -23,13 +23,13 @@ IMAGE_SIZE = 28
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
+    def inputs(self):
         """
         Define all the inputs (with type, shape, name) that
         the graph will need.
         """
-        return [InputDesc(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+        return [tf.placeholder(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
         """This function should build the model which takes the input variables

examples/basics/mnist-tfslim.py

@@ -22,9 +22,9 @@ IMAGE_SIZE = 28
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/basics/mnist-visualizations.py

@@ -68,9 +68,9 @@ def visualize_conv_activations(activation, name):
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
 

examples/basics/svhn-digit-convnet.py

@@ -22,9 +22,9 @@ Speed is about 43 it/s on TitanX.
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, [None, 40, 40, 3], 'input'),
-                InputDesc(tf.int32, [None], 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, [None, 40, 40, 3], 'input'),
+                tf.placeholder(tf.int32, [None], 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/boilerplate.py

@@ -19,9 +19,9 @@ CHANNELS = 3
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, SHAPE, SHAPE, CHANNELS), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, SHAPE, SHAPE, CHANNELS), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

examples/keras/mnist-keras.py

@@ -40,9 +40,9 @@ def get_keras_model():
 
 
 class Model(ModelDesc):
-    def _get_inputs(self):
-        return [InputDesc(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
-                InputDesc(tf.int32, (None,), 'label')]
+    def inputs(self):
+        return [tf.placeholder(tf.float32, (None, IMAGE_SIZE, IMAGE_SIZE), 'input'),
+                tf.placeholder(tf.int32, (None,), 'label')]
 
     def _build_graph(self, inputs):
         image, label = inputs

tensorpack/graph_builder/model_desc.py

@@ -102,8 +102,10 @@ class ModelDescBase(object):
         try:
             return self._get_inputs()
         except NotImplementedError:
-            with tf.Graph().as_default():   # create these placeholder in a temporary graph
+            with tf.Graph().as_default() as G:   # create these placeholder in a temporary graph
                 inputs = self.inputs()
+                for p in inputs:
+                    assert p.graph == G, "Placeholders returned by inputs() sholud be created inside inputs()!"
                 return [InputDesc.from_placeholder(p) for p in inputs]
 
     def _get_inputs(self):
@@ -117,7 +119,11 @@ class ModelDescBase(object):
         """
         __Create__ and returns a list of placeholders.
         To be implemented by subclass.
-        The placeholders __have to__ be created inside this function.
+
+        The placeholders __have to__ be created inside this method.
+        Don't return placeholders created in other methods.
+
+        You should not call this method by yourself.
 
         Returns:
             a list of `tf.placeholder`, to be converted to :class:`InputDesc`.