dynamic shape in deconv2d

README.md

@@ -2,15 +2,15 @@
 Neural Network Toolbox on TensorFlow
 
 See some [examples](examples) to learn about the framework.
-You can actually train them and reproduce the performance... not just to see how to write code.
+You can train them and reproduce the performance... not just to see how to write code.
 
 + [DoReFa-Net: training binary / low bitwidth CNN on ImageNet](examples/DoReFa-Net)
 + [ResNet for ImageNet/Cifar10/SVHN classification](examples/ResNet)
 + [InceptionV3 on ImageNet](examples/Inception/inceptionv3.py)
-+ [Fully-convolutional Network for Holistically-Nested Edge Detection](examples/HED)
-+ [Spatial Transformer Networks on MNIST addition](examples/SpatialTransformer)
-+ [Generative Adversarial Networks(GAN) variants](examples/GAN)
-+ [DQN variants on Atari games](examples/Atari2600)
++ [Fully-convolutional Network for Holistically-Nested Edge Detection(HED)](examples/HED)
++ [Spatial Transformer Network on MNIST addition](examples/SpatialTransformer)
++ [Generative Adversarial Network(GAN) variants](examples/GAN)
++ [Deep Q-Network(DQN) variants on Atari games](examples/Atari2600)
 + [Asynchronous Advantage Actor-Critic(A3C) with demos on OpenAI Gym](examples/OpenAIGym)
 + [char-rnn language model](examples/char-rnn)
 
@@ -20,13 +20,13 @@ Describe your training task with three components:
 
 1. __Model__, or graph. `models/` has some scoped abstraction of common models, but you can simply use
 	 any symbolic functions available in tensorflow, or most functions in slim/tflearn/tensorlayer.
-	`LinearWrap` and `argscope` makes large models look simpler ([vgg example](https://github.com/ppwwyyxx/tensorpack/blob/master/examples/load-vgg16.py)).
+	`LinearWrap` and `argscope` simplify large models ([vgg example](https://github.com/ppwwyyxx/tensorpack/blob/master/examples/load-vgg16.py)).
 
 2. __DataFlow__. tensorpack allows and encourages complex data processing.
 
 	+ All data producer has an unified `generator` interface, allowing them to be composed to perform complex preprocessing.
-	+ Use Python to easily handle any data format, yet still keep a good training speed thanks to multiprocess prefetch & TF Queue prefetch.
-	For example, InceptionV3 can run in the same speed as the official code which reads data using TF operators.
+	+ Use Python to easily handle any data format, yet still keep good performance thanks to multiprocess prefetch & TF Queue prefetch.
+	For example, InceptionV3 can run in the same speed as the official code which reads data by TF operators.
 
 3. __Callbacks__, including everything you want to do apart from the training iterations, such as:
 	+ Change hyperparameters during training

examples/GAN/DCGAN-CelebA.py

@@ -126,30 +126,29 @@ def sample(model_path):
         o = o[:,:,:,::-1]
         viz = next(build_patch_list(o, nr_row=10, nr_col=10, viz=True))
 
-def vec(model_path):
-    func = OfflinePredictor(PredictConfig(
-       session_init=get_model_loader(model_path),
-       model=Model(),
-       input_names=['z'],
-       output_names=['gen/gen']))
-    dic = np.load('demo/CelebA-vec.npy').item()
-    assert np.all(
-            dic['w_smile'] - dic['w_neutral'] \
-                    + dic['m_neutral'] == dic['m_smile'])
-    imgs = []
-    for z in ['w_neutral', 'w_smile', 'm_neutral', 'm_smile']:
-        z = dic[z]
-        img = func([[z]])[0][0][:,:,::-1]
-        img = (img + 1) * 128
-        imgs.append(img)
-    viz = next(build_patch_list(imgs, nr_row=1, nr_col=4, viz=True))
+#def vec(model_path):
+    #func = OfflinePredictor(PredictConfig(
+       #session_init=get_model_loader(model_path),
+       #model=Model(),
+       #input_names=['z'],
+       #output_names=['gen/gen']))
+    #dic = np.load('demo/CelebA-vec.npy').item()
+    #assert np.all(
+            #dic['w_smile'] - dic['w_neutral'] \
+                    #+ dic['m_neutral'] == dic['m_smile'])
+    #imgs = []
+    #for z in ['w_neutral', 'w_smile', 'm_neutral', 'm_smile']:
+        #z = dic[z]
+        #img = func([[z]])[0][0][:,:,::-1]
+        #img = (img + 1) * 128
+        #imgs.append(img)
+    #viz = next(build_patch_list(imgs, nr_row=1, nr_col=4, viz=True))
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
     parser.add_argument('--gpu', help='comma separated list of GPU(s) to use.')
     parser.add_argument('--load', help='load model')
     parser.add_argument('--sample', action='store_true', help='run sampling')
-    parser.add_argument('--vec', action='store_true', help='run vec arithmetic demo')
     parser.add_argument('--data', help='`image_align_celeba` directory of the celebA dataset')
     args = parser.parse_args()
     use_global_argument(args)
@@ -157,8 +156,6 @@ if __name__ == '__main__':
         os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
     if args.sample:
         sample(args.load)
-    elif args.vec:
-        vec(args.load)
     else:
         assert args.data
         config = get_config()

examples/load-alexnet.py

@@ -3,6 +3,7 @@
 # File: load-alexnet.py
 # Author: Yuxin Wu <ppwwyyxxc@gmail.com>
 
+from __future__ import print_function
 import tensorflow as tf
 import numpy as np
 import os, cv2, argparse

tensorpack/dataflow/common.py

@@ -21,7 +21,7 @@ class TestDataSpeed(ProxyDataFlow):
         self.test_size = size
 
     def get_data(self):
-        with get_tqdm(total=range(self.test_size)) as pbar:
+        with get_tqdm(total=self.test_size) as pbar:
             for dp in self.ds.get_data():
                 pbar.update()
         for dp in self.ds.get_data():

tensorpack/models/conv2d.py

@@ -34,7 +34,7 @@ def Conv2D(x, out_channel, kernel_shape,
     """
     in_shape = x.get_shape().as_list()
     in_channel = in_shape[-1]
-    assert in_channel is not None, "Input to Conv2D cannot have unknown channel!"
+    assert in_channel is not None, "[Conv2D] Input cannot have unknown channel!"
     assert in_channel % split == 0
     assert out_channel % split == 0
 
@@ -65,6 +65,26 @@ def Conv2D(x, out_channel, kernel_shape,
         nl = tf.nn.relu
     return nl(tf.nn.bias_add(conv, b) if use_bias else conv, name='output')
 
+class StaticDynamicShape(object):
+    def __init__(self, static, dynamic):
+        self.static = static
+        self.dynamic = dynamic
+
+    def apply_dynamic(self, f):
+        try:
+            return f(self.static)
+        except:
+            return f(self.dynamic)
+
+    def apply_static(self, f):
+        try:
+            return f(self.static)
+        except:
+            return None
+
+    def apply(self, f):
+        return StaticDynamicShape(self.apply_static(f), self.apply_dynamic(f))
+
 @layer_register()
 def Deconv2D(x, out_shape, kernel_shape,
            stride, padding='SAME',
@@ -86,8 +106,8 @@ def Deconv2D(x, out_shape, kernel_shape,
     :returns: a NHWC tensor
     """
     in_shape = x.get_shape().as_list()[1:]
-    assert None not in in_shape, "Input to Deconv2D cannot have unknown shape!"
     in_channel = in_shape[-1]
+    assert in_channel is not None, "[Deconv2D] Input cannot have unknown channel!"
     kernel_shape = shape2d(kernel_shape)
     stride2d = shape2d(stride)
     stride4d = shape4d(stride)
@@ -95,10 +115,16 @@ def Deconv2D(x, out_shape, kernel_shape,
 
     if isinstance(out_shape, int):
         out_channel = out_shape
-        shape3 = [stride2d[0] * in_shape[0], stride2d[1] * in_shape[1], out_shape]
+        shp3_0 = StaticDynamicShape(in_shape[0], tf.shape(x)[1]).apply(lambda x: stride2d[0] * x)
+        shp3_1 = StaticDynamicShape(in_shape[1], tf.shape(x)[2]).apply(lambda x: stride2d[1] * x)
+        shp3_dyn = [shp3_0.dynamic, shp3_1.dynamic, out_channel]
+        shp3_static = [shp3_0.static, shp3_1.static, out_channel]
     else:
+        for k in out_shape:
+            if not isinstance(k, int):
+                raise ValueError("[Deconv2D] out_shape is invalid!")
         out_channel = out_shape[-1]
-        shape3 = out_shape
+        shp3_static = shp3_dyn = out_shape
     filter_shape = kernel_shape + [out_channel, in_channel]
 
     if W_init is None:
@@ -109,7 +135,7 @@ def Deconv2D(x, out_shape, kernel_shape,
     if use_bias:
         b = tf.get_variable('b', [out_channel], initializer=b_init)
 
-    out_shape = tf.pack([tf.shape(x)[0]] + shape3)
-    conv = tf.nn.conv2d_transpose(x, W, out_shape, stride4d, padding=padding)
-    conv.set_shape(tf.TensorShape([None] + shape3))
+    out_shape_dyn = tf.pack([tf.shape(x)[0]] + shp3_dyn)
+    conv = tf.nn.conv2d_transpose(x, W, out_shape_dyn, stride4d, padding=padding)
+    conv.set_shape(tf.TensorShape([None] + shp3_static))
     return nl(tf.nn.bias_add(conv, b) if use_bias else conv, name='output')

tensorpack/tfutils/sessinit.py

@@ -194,6 +194,7 @@ def get_model_loader(filename):
     Get a corresponding model loader by looking at the file name
     :return: either a ParamRestore or SaverRestore
     """
+    assert os.path.isfile(filename), filename
     if filename.endswith('.npy'):
         return ParamRestore(np.load(filename, encoding='latin1').item())
     else:

tensorpack/utils/viz.py

@@ -144,7 +144,7 @@ def build_patch_list(patch_list,
         start = end
 
 def dump_dataflow_images(df, index=0, batched=True,
-        number=300, output_dir=None,
+        number=1000, output_dir=None,
         scale=1, resize=None, viz=None,
         flipRGB=False, exit_after=True):
     """