def prepare_logs(self, losses, input_images, output_images):
"""Return a log dictionary with all insteresting data to log.
Args:
losses (dict): A dictionary containing all important losses and skalars to log.
input_images (numpy.ndarray, torch.Tensor): Input images to log.
output_images (numpy.ndarray, torch.Tensor): Output images to log.
Returns:
dict: A dictionary containing scalars and images in a Numpy formats.
"""
logs = {"images": {}, "scalars": {**losses}}
# input images
input_img = pt2np(input_images)
logs["images"].update({"batch_input": input_img})
# output images
output_img = pt2np(output_images)
logs["images"].update({"batch_output": output_img})
# log only max three images separately
max_num = 3 if self.config["batch_size"] > 3 else self.config[
"batch_size"]
for i in range(max_num):
logs["images"].update(
{"input_" + str(i): np.expand_dims(input_img[i], 0)})
logs["images"].update(
{"output_" + str(i): np.expand_dims(output_img[i], 0)})
logs = convert_logs2numpy(logs)
return logs
def prepare_logs(self, losses, predictions):
"""Return a log dictionary with all instersting data to log."""
# create a dictionary to log with all interesting variables
logs = {"images": {}, "scalars": {**losses}}
# generated images
output_img = pt2np(predictions)
logs["images"].update({"batch_output": output_img})
# log only max three images separately
max_num = 3 if self.config["batch_size"] > 3 else self.config[
"batch_size"]
for i in range(max_num):
logs["images"].update(
{"output_" + str(i): np.expand_dims(output_img[i], 0)})
# convert to numpy
logs = convert_logs2numpy(logs)
return logs
def prepare_logs(self, losses, inputs, predictions):
"""Return a log dictionary with all instersting data to log."""
# create a dictionary to log with all interesting variables
logs = {"images": {}, "scalars": {**losses}}
############
## images ##
############
# input images
real_A_img = pt2np(inputs[0])
real_B_img = pt2np(inputs[1])
logs["images"].update({"batch_input_sketch": real_A_img})
logs["images"].update({"batch_input_face": real_B_img})
# fake images
fake_A_img = pt2np(predictions[1])
fake_B_img = pt2np(predictions[0])
logs["images"].update({"batch_fake_sketch": fake_A_img})
logs["images"].update({"batch_fake_face": fake_B_img})
# reconstructed images
rec_A_img = pt2np(self.model.output['rec_A'])
rec_B_img = pt2np(self.model.output['rec_B'])
logs["images"].update({"batch_rec_sketch": rec_A_img})
logs["images"].update({"batch_rec_face": rec_B_img})
# log only max three images separately
max_num = 3 if self.config["batch_size"] > 3 else self.config[
"batch_size"]
for i in range(max_num):
logs["images"].update(
{"input_sketch_" + str(i): np.expand_dims(real_A_img[i], 0)})
logs["images"].update(
{"input_face_" + str(i): np.expand_dims(real_B_img[i], 0)})
logs["images"].update(
{"fake_sketch_" + str(i): np.expand_dims(fake_A_img[i], 0)})
logs["images"].update(
{"fake_face_" + str(i): np.expand_dims(fake_B_img[i], 0)})
logs["images"].update(
{"rec_sketch_" + str(i): np.expand_dims(rec_A_img[i], 0)})
logs["images"].update(
{"rec_face_" + str(i): np.expand_dims(rec_B_img[i], 0)})
# convert to numpy
logs = convert_logs2numpy(logs)
return logs
def prepare_logs(self, losses, inputs, predictions):
"""Return a log dictionary with all insteresting data to log.
Args:
losses (dict): A dictionary containing all important losses and skalars to log.
inputs (numpy.ndarray, torch.Tensor): Input images to log.
predictions (numpy.ndarray, torch.Tensor): Output images to log.
Returns:
dict: A dictionary containing scalars and images in a Numpy formats.
"""
logs = {"images": {}, "scalars": {**losses}}
# input images
real_A_img = pt2np(inputs[0])
real_B_img = pt2np(inputs[1])
logs["images"].update({"batch_input_sketch": real_A_img})
logs["images"].update({"batch_input_face": real_B_img})
# fake images
fake_A_img = pt2np(predictions[1])
fake_B_img = pt2np(predictions[0])
logs["images"].update({"batch_fake_sketch": fake_A_img})
logs["images"].update({"batch_fake_face": fake_B_img})
# reconstruction images
rec_A_img = pt2np(self.model.output['rec_A'])
rec_B_img = pt2np(self.model.output['rec_B'])
logs["images"].update({"batch_rec_sketch": rec_A_img})
logs["images"].update({"batch_rec_face": rec_B_img})
# log only max three images separately
max_num = 3 if self.batch_size > 3 else self.batch_size
for i in range(max_num):
logs["images"].update(
{"input_sketch_" + str(i): np.expand_dims(real_A_img[i], 0)})
logs["images"].update(
{"input_face_" + str(i): np.expand_dims(real_B_img[i], 0)})
logs["images"].update(
{"fake_sketch_" + str(i): np.expand_dims(fake_A_img[i], 0)})
logs["images"].update(
{"fake_face_" + str(i): np.expand_dims(fake_B_img[i], 0)})
logs = convert_logs2numpy(logs)
return logs