def visualize_result_on_samples(epoch,
model,
sample_images,
logger,
step,
split='train'):
model.eval()
with torch.no_grad():
sample_images = sample_images.to(device=device, dtype=dtype)
scores = model(sample_images).cpu().numpy()
images_li = []
for i in range(scores.shape[0]):
input = scores[i, :, :, :].squeeze()
picture = render(input)
if i == 0:
truth = sample_images[i, :, :, :].cpu().numpy().squeeze()
truth = np.moveaxis(truth, 0, -1)
toprint = np.moveaxis(picture, 0, -1)
io.imsave(
'img/prediction_epoch{}_step{}.png'.format(epoch, step),
toprint)
io.imsave('img/truth_epoch{}_step{}.png'.format(epoch, step),
truth)
images_li.append(picture)
logger.image_summary('result_{}'.format(split), images_li, step)
def visualize_result_on_samples(model, sample_images, logger, step, split='train'):
model.eval()
with torch.no_grad():
sample_images = sample_images.to(device=device, dtype=dtype)
scores = model(sample_images).cpu().numpy()
images_li = []
for i in range(scores.shape[0]):
input = scores[i, :, :, :].squeeze()
picture = render(input)
images_li.append(picture)
logger.image_summary('result_{}'.format(split), images_li, step)