Add save_summaries option, fix typos

README.md

@@ -15,7 +15,6 @@ Project Organization
  ├── Makefile <- Makefile with commands like `make data` or `make train`
  ├── setup.py <- makes project pip installable (pip install -e .) so src can be imported
  ├── LICENSE
- ├── eval_cityscapes <- This contains scripts for getting the experiment results.
  └── src <- Source code for use in this project
      ├── __init__.py <- Makes src a Python module
   │

src/data/dataset.py

@@ -54,7 +54,7 @@ def load_train_data(self):
  if self.opt.gpu_id != -1:
  train_datasetA = train_datasetA.apply(tf.contrib.data.prefetch_to_device(self.gpu_id, buffer_size=1))
  train_datasetB = train_datasetB.apply(tf.contrib.data.prefetch_to_device(self.gpu_id, buffer_size=1))
- # Create a tf.data.Iterator from the Datasets:
+ # Create a tf.data.Iterator from each Dataset:
  return iter(train_datasetA), iter(train_datasetB)
 
  def load_test_data(self):

src/models/networks.py

@@ -233,5 +233,4 @@ def call(self, inputs):
  x = self.leaky(x)
 
  x = self.conv5(x)
- #x = tf.nn.sigmoid(x) # use_sigmoid = not use_lsgan TODO
  return x

src/test.py

@@ -10,7 +10,7 @@
 tf.enable_eager_execution()
 """
 Run this module for testing.
-Required args: --data_dir, --save_dir
+Required args: --data_dir, --save_dir, --results_dir
 """
 if __name__ == "__main__":
  opt = Options().parse(training=False)

src/train.py

@@ -10,14 +10,15 @@
 tf.enable_eager_execution()
 """
 Run this module for training.
-Required args: --data_dir, --save_dir, --results_dir
+Required args: --data_dir, --save_dir
 """
 if __name__ == "__main__":
  opt = Options().parse(training=True)
  dataset = Dataset(opt)
  model = CycleGANModel(opt)
 
  device = ("/gpu:" + str(opt.gpu_id)) if opt.gpu_id != -1 else "/cpu:0"
+
  with tf.device(device):
  global_step = model.global_step
  batches_per_epoch = dataset.get_batches_per_epoch(opt)
@@ -26,24 +27,26 @@
  summary_writer = tf.contrib.summary.create_file_writer(log_dir, flush_millis=10000)
  for epoch in range(1, opt.epochs):
  start = time.time()
- with summary_writer.as_default():
  for train_step in range(batches_per_epoch):
  # Record summaries every 100 train_steps, we multiply by 3 because there are 3 gradient updates per step.
- with tf.contrib.summary.record_summaries_every_n_global_steps(opt.summary_freq * 3, global_step=global_step):
+ with summary_writer.as_default(),
+ tf.contrib.summary.record_summaries_every_n_global_steps(opt.summary_freq * 3, global_step=global_step):
  model.set_input(dataset.data)
  model.optimize_parameters()
- # Summaries for Tensorboard:
- tf.contrib.summary.scalar('loss/genA2B', model.genA2B_loss)
- tf.contrib.summary.scalar('loss/genB2A', model.genB2A_loss)
- tf.contrib.summary.scalar('loss/discA', model.discA_loss)
- tf.contrib.summary.scalar('loss/discB', model.discB_loss)
- tf.contrib.summary.scalar('loss/cyc', model.cyc_lossA + model.cyc_lossB)
- tf.contrib.summary.scalar('loss/identity', model.id_lossA + model.id_lossB)
- tf.contrib.summary.scalar('learning_rate', model.learning_rate)
- tf.contrib.summary.image('A/generated', model.fakeA)
- tf.contrib.summary.image('A/reconstructed', model.reconstructedA)
- tf.contrib.summary.image('B/generated', model.fakeB)
- tf.contrib.summary.image('B/reconstructed', model.reconstructedB)
+ if opt.save_summaries:
+ # Summaries for Tensorboard:
+ tf.contrib.summary.scalar('loss/genA2B', model.genA2B_loss)
+ tf.contrib.summary.scalar('loss/genB2A', model.genB2A_loss)
+ tf.contrib.summary.scalar('loss/discA', model.discA_loss)
+ tf.contrib.summary.scalar('loss/discB', model.discB_loss)
+ tf.contrib.summary.scalar('loss/cyc', model.cyc_lossA + model.cyc_lossB)
+ tf.contrib.summary.scalar('loss/identity', model.id_lossA + model.id_lossB)
+ tf.contrib.summary.scalar('learning_rate', model.learning_rate)
+ tf.contrib.summary.image('A/generated', model.fakeA)
+ tf.contrib.summary.image('A/reconstructed', model.reconstructedA)
+ tf.contrib.summary.image('B/generated', model.fakeB)
+ tf.contrib.summary.image('B/reconstructed', model.reconstructedB)
+ print("Iteration complete")
  # Assign decayed learning rate:
  model.update_learning_rate(batches_per_epoch)
  # Checkpoint the model:

src/utils/options.py

@@ -19,7 +19,7 @@ def __init__(self):
  parser.add_argument('--instance_norm', action='store_false', help='if true, uses instance normalisation after each conv layer in D and G')
  parser.add_argument('--init_scale', type=float, default=0.02, help='stddev for weight initialisation; small variance helps prevent colour inversion.')
  parser.add_argument('--gen_skip', action='store_true', help='if true, use skip connection from first residual block to last in generator')
- parser.add_argument('--resize_conv', action='store_true', help='if true, replace conv2dtranspose in generator with upsample -> conv2d')
+ parser.add_argument('--resize_conv', action='store_false', help='if true, replace conv2dtranspose in generator with upsample -> conv2d')
  parser.add_argument('--use_dropout', action='store_true', help='if true, use dropout for the generator')
  parser.add_argument('--dropout_prob', type=float, default=0.5, help='dropout probability for all layers in generator')
  # dataset options
@@ -56,6 +56,7 @@ def _get_train_options(self, parser):
  parser.add_argument('--training', action='store_false', help='boolean for training/testing')
  parser.add_argument('--load_checkpoint', action='store_false', help='if true, loads latest checkpoint')
  parser.add_argument('--save_epoch_freq', type=int, default=5, help='frequency of saving checkpoints at the end of epochs')
+ parser.add_argument('--save_summaries', action='store_false', help='if true, stores tensorboard summaries. Turn off to speed up training significantly')
  parser.add_argument('--summary_freq', type=int, default=100, help='frequency of saving saving tensorboard summaries in training steps')
  parser.add_argument('--epochs', type=int, default=200, help='number of epochs to train the model; learning rate decays to 0 by epoch 200')
  parser.add_argument('--batch_size', type=int, default=1, help='input batch size')