def _visualize(self, info, logger):
N = 16
real = info.real[:N // 2]
logger.add('images', 'real-img', util.image_size_limiter(real))
gen = info.gen[:N]
logger.add('images', 'gen-img', util.image_size_limiter(gen))
def _visualize(self, info, records):
settings = self._viz_settings
if 'gen' in settings and isinstance(self.generator, learn.Visualizable):
self.generator.visualize(info, records)
if 'disc' in settings and isinstance(self.discriminator, learn.Visualizable):
self.discriminator.visualize(info, records)
if 'samples' in settings:
N = 16
real = info.real[:N // 2]
records.log('images', 'real-img', util.image_size_limiter(real))
gen = info.gen[:N]
records.log('images', 'gen-img', util.image_size_limiter(gen))
def _visualize(self, info, records):
settings = self._viz_settings
super()._visualize(info, records)
B, C, *other = info.original.size()
N = min(B, 8)
if len(other) and ('gen-prior' in self._viz_settings or not self.training):
viz_gen = self.generate_prior(2 * N).view(2*N, C, *other)
if 'force' in settings:
viz_gen = viz_gen.sigmoid()
records.log('images', 'gen-prior', util.image_size_limiter(viz_gen))
def _visualize(self, info, records):
N = 8
# real = torch.cat([info.real_A[:N], info.real_B[:N]])
# logger.add('images', 'real-img', util.image_size_limiter(real))
real_A = info.real_A[:N]
records.log('images', 'real-img-A', util.image_size_limiter(real_A))
real_B = info.real_B[:N]
records.log('images', 'real-img-B', util.image_size_limiter(real_B))
gen_A = info.fake_A[:N * 2]
records.log('images', 'gen-img-A', util.image_size_limiter(gen_A))
gen_B = info.fake_B[:N * 2]
records.log('images', 'gen-img-B', util.image_size_limiter(gen_B))
rec_A = info.recov_A[:N * 2]
records.log('images', 'rec-img-A', util.image_size_limiter(rec_A))
rec_B = info.recov_B[:N * 2]
records.log('images', 'rec-img-B', util.image_size_limiter(rec_B))