simplify (and fix) some examples

examples/Char-RNN/char-rnn.py

@@ -116,7 +116,6 @@ def get_config():
 
     ds = CharRNNData(param.corpus, 100000)
     ds = BatchData(ds, param.batch_size)
-    steps_per_epoch = ds.size()
 
     return TrainConfig(
         dataflow=ds,
@@ -125,12 +124,9 @@ def get_config():
             ScheduledHyperParamSetter('learning_rate', [(25, 2e-4)])
         ],
         model=Model(),
-        steps_per_epoch=steps_per_epoch,
         max_epoch=50,
     )
 
-# TODO rewrite using Predictor interface
-
 
 def sample(path, start, length):
     """

examples/GAN/Image2Image.py

@@ -158,7 +158,6 @@ def get_data():
     imgs = glob.glob(os.path.join(datadir, '*.jpg'))
     ds = ImageFromFile(imgs, channel=3, shuffle=True)
 
-    # Image-to-Image translation mode
     ds = MapData(ds, lambda dp: split_input(dp[0]))
     assert SHAPE < 286  # this is the parameter used in the paper
     augs = [imgaug.Resize(286), imgaug.RandomCrop(SHAPE)]

examples/GAN/InfoGAN-mnist.py

@@ -151,7 +151,7 @@ class Model(GANModelDesc):
         opt = tf.train.AdamOptimizer(lr, beta1=0.5, epsilon=1e-6)
         # generator learns 5 times faster
         return optimizer.apply_grad_processors(
-            opt, [gradproc.ScaleGradient(('.*', 5), log=False)])
+            opt, [gradproc.ScaleGradient(('gen/.*', 5), log=True)])
 
 
 def get_data():

examples/Inception/inception-bn.py

@@ -156,9 +156,7 @@ def get_data(train_or_test):
 
 def get_config():
     logger.auto_set_dir()
-    # prepare dataset
     dataset_train = get_data('train')
-    steps_per_epoch = 5000
     dataset_val = get_data('val')
 
     return TrainConfig(
@@ -175,7 +173,7 @@ def get_config():
         ],
         session_config=get_default_sess_config(0.99),
         model=Model(),
-        steps_per_epoch=steps_per_epoch,
+        steps_per_epoch=5000,
         max_epoch=80,
     )
 

examples/ResNet/cifar10-resnet.py

@@ -139,10 +139,7 @@ def get_data(train_or_test):
 
 def get_config():
     logger.auto_set_dir()
-
-    # prepare dataset
     dataset_train = get_data('train')
-    steps_per_epoch = dataset_train.size()
     dataset_test = get_data('test')
 
     return TrainConfig(
@@ -155,7 +152,6 @@ def get_config():
                                       [(1, 0.1), (82, 0.01), (123, 0.001), (300, 0.0002)])
         ],
         model=Model(n=NUM_UNITS),
-        steps_per_epoch=steps_per_epoch,
         max_epoch=400,
     )
 

examples/SimilarityLearning/mnist-embeddings.py

@@ -137,15 +137,12 @@ class SoftTripletModel(TripletModel):
 def get_config(model, algorithm_name):
     logger.auto_set_dir()
 
-    dataset = model.get_data()
-    steps_per_epoch = dataset.size()
-
     extra_display = ["cost"]
     if not algorithm_name == "cosine":
         extra_display = extra_display + ["loss/pos-dist", "loss/neg-dist"]
 
     return TrainConfig(
-        dataflow=dataset,
+        dataflow=model.get_data(),
         model=model(),
         callbacks=[
             ModelSaver(),
@@ -155,7 +152,6 @@ def get_config(model, algorithm_name):
             MovingAverageSummary(),
             ProgressBar(extra_display),
             StatPrinter()],
-        steps_per_epoch=steps_per_epoch,
         max_epoch=20,
     )
 

examples/SpatialTransformer/mnist-addition.py

@@ -154,6 +154,7 @@ def get_config():
     steps_per_epoch = dataset_train.size() * 5
 
     return TrainConfig(
+        model=Model(),
         dataflow=dataset_train,
         callbacks=[
             ModelSaver(),
@@ -162,7 +163,6 @@ def get_config():
             ScheduledHyperParamSetter('learning_rate', [(200, 1e-4)])
         ],
         session_config=get_default_sess_config(0.5),
-        model=Model(),
         steps_per_epoch=steps_per_epoch,
         max_epoch=500,
     )

examples/cifar-convnet.py

@@ -110,7 +110,6 @@ def get_config(cifar_classnum):
 
     # prepare dataset
     dataset_train = get_data('train', cifar_classnum)
-    steps_per_epoch = dataset_train.size()
     dataset_test = get_data('test', cifar_classnum)
 
     sess_config = get_default_sess_config(0.5)
@@ -120,6 +119,7 @@ def get_config(cifar_classnum):
             raise StopTraining()
         return lr * 0.31
     return TrainConfig(
+        model=Model(cifar_classnum),
         dataflow=dataset_train,
         callbacks=[
             ModelSaver(),
@@ -128,8 +128,6 @@ def get_config(cifar_classnum):
                                    threshold=0.001, last_k=10),
         ],
         session_config=sess_config,
-        model=Model(cifar_classnum),
-        steps_per_epoch=steps_per_epoch,
         max_epoch=150,
     )
 

examples/mnist-convnet.py

@@ -140,6 +140,7 @@ def get_config():
 
     # get the config which contains everything necessary in a training
     return TrainConfig(
+        model=Model(),
         dataflow=dataset_train,  # the DataFlow instance for training
         callbacks=[
             ModelSaver(),   # save the model after every epoch
@@ -148,7 +149,6 @@ def get_config():
                 # Calculate both the cost and the error for this DataFlow
                 [ScalarStats('cross_entropy_loss'), ClassificationError('incorrect')]),
         ],
-        model=Model(),
         steps_per_epoch=steps_per_epoch,
         max_epoch=100,
     )

examples/svhn-digit-convnet.py

@@ -96,17 +96,15 @@ def get_data():
 def get_config():
     logger.auto_set_dir()
     data_train, data_test = get_data()
-    steps_per_epoch = data_train.size()
 
     return TrainConfig(
+        model=Model(),
         dataflow=data_train,
         callbacks=[
             ModelSaver(),
             InferenceRunner(data_test,
                             [ScalarStats('cost'), ClassificationError()])
         ],
-        model=Model(),
-        steps_per_epoch=steps_per_epoch,
         max_epoch=350,
     )
 

tensorpack/dataflow/image.py

@@ -37,7 +37,7 @@ class ImageFromFile(RNGDataFlow):
         for f in self.files:
             im = cv2.imread(f, self.imread_mode)
             if self.channel == 3:
-                im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
+                im = im[:, :, ::-1]
             if self.resize is not None:
                 im = cv2.resize(im, self.resize[::-1])
             if self.channel == 1: