small clean-ups

examples/CaffeModels/load-alexnet.py

@@ -13,7 +13,6 @@ import tensorflow as tf
 from tensorpack import *
 from tensorpack.dataflow.dataset import ILSVRCMeta
 from tensorpack.tfutils.summary import *
-from tensorpack.tfutils.symbolic_functions import *
 
 
 def tower_func(image):

examples/FasterRCNN/eval.py

@@ -234,7 +234,8 @@ class EvalCallback(Callback):
                                                 shard=k, num_shards=self.num_predictor)
                               for k in range(self.num_predictor)]
         else:
-            # Only eval on the first machine.
+            # Only eval on the first machine,
+            # Because evaluation assumes that all horovod workers share the filesystem.
             # Alternatively, can eval on all ranks and use allgather, but allgather sometimes hangs
             self._horovod_run_eval = hvd.rank() == hvd.local_rank()
             if self._horovod_run_eval:

examples/GAN/Improved-WGAN.py

@@ -70,7 +70,7 @@ class Model(DCGAN.Model):
         # the gradient penalty loss
         gradients = tf.gradients(vec_interp, [interp])[0]
         gradients = tf.sqrt(tf.reduce_sum(tf.square(gradients), [1, 2, 3]))
-        gradients_rms = symbolic_functions.rms(gradients, 'gradient_rms')
+        gradients_rms = tf.sqrt(tf.reduce_mean(tf.square(gradients)), name='gradient_rms')
         gradient_penalty = tf.reduce_mean(tf.square(gradients - 1), name='gradient_penalty')
         add_moving_summary(self.d_loss, self.g_loss, gradient_penalty, gradients_rms)
 

examples/GAN/README.md

 # Generative Adversarial Networks
 
-Reproduce the following GAN-related methods, 100~200 lines each:
+Reproduce the following GAN-related methods, __100~200 lines each__:
 
 + DCGAN ([Unsupervised Representation Learning with DCGAN](https://arxiv.org/abs/1511.06434))
 

examples/OpticalFlow/flownet2.py

@@ -28,7 +28,7 @@ def apply(model, model_path, images, ground_truth=None):
         input_names=['left', 'right'],
         output_names=['prediction']))
 
-    for right in images[1:]:
+    for idx, right in enumerate(images[1:]):
         right = aug.augment(cv2.imread(right))
 
         left_input, right_input = [x.astype('float32').transpose(2, 0, 1)[None, ...]
@@ -43,7 +43,7 @@ def apply(model, model_path, images, ground_truth=None):
         img = viz.stack_patches(patches, 2, 2)
 
         cv2.imshow('flow output', img)
-        cv2.imwrite('flow_prediction.png', img)
+        cv2.imwrite('flow_output{:03d}.png'.format(idx), img)
         cv2.waitKey(0)
 
         left = right

examples/Saliency/CAM-resnet.py

@@ -14,7 +14,6 @@ from tensorpack import *
 from tensorpack.dataflow import dataset
 from tensorpack.tfutils import gradproc, optimizer
 from tensorpack.tfutils.summary import *
-from tensorpack.tfutils.symbolic_functions import *
 from tensorpack.utils import viz
 from tensorpack.utils.gpu import get_num_gpu