def save_stats(self, epoch):
# Save logger
if epoch is not None:
# add log
self.logger_stats.write(
'----------------- Epoch scores summary ------------------------- \n'
)
self.logger_stats.write(
'[epoch %d], [val loss %.5f], [acc %.2f], [mean_IoU %.2f] \n'
% (epoch, self.stats.val.loss, 100 * self.stats.val.acc,
100 * self.stats.val.mIoU))
self.logger_stats.write(
'---------------------------------------------------------------- \n'
)
# add scores to tensorboard
self.writer.add_scalar('losses/epoch', self.stats.val.loss,
epoch)
self.writer.add_scalar('metrics/accuracy',
100. * self.stats.val.acc, epoch)
self.writer.add_scalar('metrics/mIoU',
100. * self.stats.val.mIoU, epoch)
conf_mat_img = confm_metrics2image(
self.stats.val.get_confm_norm(), self.cf.labels)
#self.writer.add_image('metrics/conf_matrix', conf_mat_img, epoch)
else:
self.logger_stats.write(
'----------------- Scores summary -------------------- \n')
self.logger_stats.write(
'[val loss %.5f], [acc %.2f], [mean_IoU %.2f]\n' %
(self.stats.val.loss, 100 * self.stats.val.acc,
100 * self.stats.val.mIoU))
self.logger_stats.write(
'---------------------------------------------------------------- \n'
)
def save_stats(self, epoch, mode):
# Save logger
if epoch is not None:
# add scores to log
self.logger_stats.write('Classification validation scores:\n')
self.logger_stats.write(
'[epoch %d], [val loss %.5f], [acc %.2f], [precision %.2f], [recall %.2f], [f1score %.2f]\n' % (
epoch, self.stats.val.loss, 100 * self.stats.val.acc, 100 * self.stats.val.precision,
100 * self.stats.val.recall, 100 * self.stats.val.f1score))
# add scores to tensorboard
self.writer.add_scalar('losses/epoch', self.stats.val.loss, epoch)
# self.writer.add_scalar('metrics/accuracy', 100. * self.stats.val.acc, epoch)
self.writer.add_scalar('metrics/precision', 100. * self.stats.val.precision, epoch)
# self.writer.add_scalar('metrics/recall', 100. * self.stats.val.recall, epoch)
# self.writer.add_scalar('metrics/f1score', 100. * self.stats.val.f1score, epoch)
conf_mat_img = confm_metrics2image(self.stats.val.get_confm_norm(), self.cf.labels)
self.writer.add_image('metrics/conf_matrix', conf_mat_img, epoch, dataformats='HWC')
else:
self.logger_stats.write('----------------- Scores summary --------------------\n')
self.logger_stats.write(
'[%s loss %.5f], [acc %.2f], [precision %.2f], [recall %.2f], [f1score %.2f]\n' % (
mode, self.stats.val.loss, 100 * self.stats.val.acc, 100 * self.stats.val.precision,
100 * self.stats.val.recall, 100 * self.stats.val.f1score))
self.logger_stats.write('---------------------------------------------------------------- \n')
def save_stats_epoch(self, epoch):
# Save logger
if epoch is not None:
# Epoch loss tensorboard
self.writer.add_scalar('losses/epoch', self.stats.train.loss, epoch)
self.writer.add_scalar('metrics/accuracy', 100.*self.stats.train.acc, epoch)
self.writer.add_scalar('metrics/mIoU', 100.*self.stats.train.mIoU, epoch)
conf_mat_img = confm_metrics2image(self.stats.train.get_confm_norm(), self.cf.labels)
def save_stats_epoch(self, epoch):
# Save logger
if epoch is not None:
# Epoch loss tensorboard
self.writer.add_scalar('losses/epoch', self.stats.train.loss, epoch)
# self.writer.add_scalar('metrics/accuracy', 100. * self.stats.train.acc, epoch)
self.writer.add_scalar('metrics/precision', 100. * self.stats.train.precision, epoch)
# self.writer.add_scalar('metrics/recall', 100. * self.stats.train.recall, epoch)
# self.writer.add_scalar('metrics/f1score', 100. * self.stats.train.f1score, epoch)
conf_mat_img = confm_metrics2image(self.stats.train.get_confm_norm(), self.cf.labels)
self.writer.add_image('metrics/conf_matrix', conf_mat_img, epoch, dataformats='HWC')
# Save learning rate
# self.logger_stats.write(str(self.model.scheduler.get_lr())) # Step, MultiStep
for param_group in self.model.optimizer.param_groups:
self.writer.add_scalar('lr/lr', param_group['lr'], epoch) # ReduceLROnPlateau
def save_stats_epoch(self, epoch):
# Save logger
if epoch is not None:
# Epoch loss tensorboard
self.writer.add_scalar('losses/epoch', self.stats.train.loss,
epoch)
self.writer.add_scalar('metrics/accuracy',
100. * self.stats.train.acc, epoch)
self.writer.add_scalar('metrics/mIoU',
100. * self.stats.train.mIoU, epoch)
conf_mat_img = confm_metrics2image(
self.stats.train.get_confm_norm(), self.cf.labels)
self.writer.add_image('metrics/conf_matrix',
conf_mat_img,
epoch,
dataformats='HWC')
for param_group in self.model.optimizer.param_groups:
self.writer.add_scalar('lr/lr', param_group['lr'], epoch)
def save_stats_epoch(self, epoch):
# Save logger
if epoch is not None:
# Epoch loss tensorboard
self.writer.add_scalar('losses/epoch', self.stats.train.loss,
epoch)
self.writer.add_scalar('metrics/accuracy',
100. * self.stats.train.acc, epoch)
self.writer.add_scalar('metrics/precision',
100. * self.stats.train.precision,
epoch)
self.writer.add_scalar('metrics/recall',
100. * self.stats.train.recall, epoch)
self.writer.add_scalar('metrics/f1score',
100. * self.stats.train.f1score, epoch)
conf_mat_img = confm_metrics2image(
self.stats.train.get_confm_norm(), self.cf.labels)
self.writer.add_image('metrics/conf_matrix', conf_mat_img,
epoch)
