def _log_fn(self, context: GANEncoderContext):
"""
Log output of the generator to Tensorboard.
Logs G(E(x)).
Args:
context (:py:class:`ashpy.contexts.gan.GanEncoderContext`): current context.
"""
if context.log_eval_mode == LogEvalMode.TEST:
generator_of_encoder = context.generator_model(
context.encoder_model(context.encoder_inputs, training=False),
training=False,
)
elif context.log_eval_mode == LogEvalMode.TRAIN:
generator_of_encoder = context.generator_of_encoder
else:
raise ValueError("Invalid LogEvalMode")
# Tensorboard 2.0 does not support float images in [-1, 1]
# Only in [0,1]
if generator_of_encoder.dtype == tf.float32:
# The hypothesis is that image are in [-1,1] how to check?
generator_of_encoder = (generator_of_encoder + 1.0) / 2
log("generator_of_encoder", generator_of_encoder, context.global_step)
def _log_fn(context: ClassifierContext) -> None:
"""
Log output of the image and label to Tensorboard.
Args:
context: current context
"""
input_tensor = context.current_batch[0]
out_label = context.current_batch[1]
log("input_x", input_tensor, context.global_step)
log("input_y", out_label, context.global_step)
def _log_fn(self, context: GANContext) -> None:
"""
Log output of the generator to Tensorboard.
Args:
context (:py:class:`ashpy.contexts.gan.GANContext`): current context.
"""
if context.log_eval_mode == LogEvalMode.TEST:
out = context.generator_model(context.generator_inputs,
training=False)
elif context.log_eval_mode == LogEvalMode.TRAIN:
out = context.fake_samples
else:
raise ValueError("Invalid LogEvalMode")
# tensorboard 2.0 does not support float images in [-1, 1]
# only in [0,1]
if out.dtype == tf.float32:
# The hypothesis is that image are in [-1,1] how to check?
out = (out + 1.0) / 2
log("generator", out, context.global_step)