def _collect_running_batch_states(self):
self.running_acc.append(self._compute_acc().item())
m = len(self.dataloaders['train'])
if self.is_training is False:
m = len(self.dataloaders['val'])
if np.mod(self.batch_id, 100) == 1 or self.batch_id == m - 1:
print(
'Is_training: %s. [%d,%d][%d,%d], G_loss: %.8f, D_loss: %.8f, running_acc: %.8f (%s),'
% (self.is_training, self.epoch_id, self.max_num_epochs - 1,
self.batch_id, m, self.G_loss.item(), self.D_loss.item(),
np.mean(self.running_acc), self.metric))
if np.mod(self.batch_id, 1000) == 1 or self.batch_id == m - 1:
vis_input = utils.make_numpy_grid(self.batch['input'])
vis_pred1 = utils.make_numpy_grid(self.G_pred1)
vis_pred2 = utils.make_numpy_grid(self.G_pred2)
if self.output_auto_enhance:
vis_pred1 = vis_pred1 * 1.5
vis_pred2 = vis_pred2 * 1.5
vis = np.concatenate([vis_input, vis_pred1, vis_pred2], axis=0)
vis = np.clip(vis, a_min=0.0, a_max=1.0)
file_name = os.path.join(
self.vis_dir, 'istrain_' + str(self.is_training) + '_' +
str(self.epoch_id) + '_' + str(self.batch_id) + '.jpg')
plt.imsave(file_name, vis)
def _collect_running_batch_states(self):
self.running_acc.append(self._compute_acc().item())
m = len(self.dataloaders['train'])
if self.is_training is False:
m = len(self.dataloaders['val'])
if np.mod(self.batch_id, 100) == 1:
print(
'Is_training: %s. [%d,%d][%d,%d], G_loss: %.5f, running_acc: %.5f'
% (self.is_training, self.epoch_id, self.max_num_epochs - 1,
self.batch_id, m, self.G_loss.item(),
np.mean(self.running_acc)))
if np.mod(self.batch_id, 1000) == 1:
vis_pred_foreground = utils.make_numpy_grid(self.G_pred_foreground)
vis_gt_foreground = utils.make_numpy_grid(self.batch['B'])
vis_pred_alpha = utils.make_numpy_grid(self.G_pred_alpha)
vis_gt_alpha = utils.make_numpy_grid(self.batch['ALPHA'])
vis = np.concatenate([
vis_pred_foreground, vis_gt_foreground, vis_pred_alpha,
vis_gt_alpha
],
axis=0)
vis = np.clip(vis, a_min=0.0, a_max=1.0)
file_name = os.path.join(
self.vis_dir, 'istrain_' + str(self.is_training) + '_' +
str(self.epoch_id) + '_' + str(self.batch_id) + '.jpg')
plt.imsave(file_name, vis)
def _visualize_batch_and_prediction(self):
if np.mod(self.batch_id, 100) == 1:
vis_input = utils.make_numpy_grid(self.batch['stereogram'])
vis_pred = utils.make_numpy_grid(self.G_pred)
vis_gt = utils.make_numpy_grid(self.batch['dmap'])
vis = np.concatenate([vis_input, vis_pred, vis_gt], axis=0)
vis = np.clip(vis, a_min=0.0, a_max=1.0)
file_name = os.path.join(
self.vis_dir, 'istrain_' + str(self.is_training) + '_' +
str(self.epoch_id) + '_' + str(self.batch_id) + '.jpg')
plt.imsave(file_name, vis)