def run1(self):
""" FGVC evaluation loop """
acc_dict = {}
bits_per_pixel = []
self.logger.info('classifiers for evaluation: {}'.format(
[m for m in self.fgvc_cnn_model_names]))
for cnn_model_name in self.fgvc_cnn_model_names:
model_checkpoint = self.get_fgvc_checkpoint(cnn_model_name)
if model_checkpoint is None:
continue
bpp, accs = self.eval_classifier_model(
cnn_model=self.get_classifier_instance(cnn_model_name),
ckpt_path=model_checkpoint)
acc_dict[cnn_model_name] = accs
bits_per_pixel.append(bpp)
# write results to csv
csv_file = os.path.join(
self._log_dir,
'{}_accuracy_original_{}.csv'.format(cnn_model_name,
self._log_number))
write_to_csv(csv_file, [(bpp, *accs)])
# log results
accuracy_dicts = [{cnn_model_name: accs}]
self.log_results(
self.pack_eval_values(bits_per_pixel=[bpp],
accuracy_dicts=accuracy_dicts))
def run0(self):
""" imagenet evaluation loop """
acc_dict = {}
bits_per_pixel = []
self.logger.info('classifiers for evaluation: {}'.format(
[m for m in self.imagenet_cnn_model_names]))
for cnn_model_name in self.imagenet_cnn_model_names:
bpp, accs = self.eval_classifier_model(
self.get_classifier_instance(cnn_model_name))
acc_dict[cnn_model_name] = accs
bits_per_pixel.append(bpp)
# write results to csv
csv_file = os.path.join(
self._log_dir,
'{}_accuracy_original_{}.csv'.format(cnn_model_name,
self._log_number))
write_to_csv(csv_file, [(bpp, *accs)])
# log results
accuracy_dicts = [{cnn_model_name: accs}]
self.log_results(
self.pack_eval_values(bits_per_pixel=[bpp],
accuracy_dicts=accuracy_dicts))
def _save_results(self, bpp, accs, cnn_model_name):
# write results to csv
csv_file = os.path.join(self._log_dir, 'cpm_{}_accuracy_{}.csv'.format(cnn_model_name, self._log_number))
write_to_csv(csv_file, [(b, *a) for b, a in zip(bpp, accs)])
# log results
accuracy_dicts = [{cnn_model_name: acc} for acc in accs]
self.log_results(self.pack_eval_values(bits_per_pixel=bpp, accuracy_dicts=accuracy_dicts))
def _save_results(self, bpp, accs, cnn_model_name):
# write results to csv
csv_file = os.path.join(
self._log_dir,
'webp_{}_accuracy_{}.csv'.format(cnn_model_name, self._log_number))
write_to_csv(csv_file, [(b, *a) for b, a in zip(bpp, accs)])
# log results
accuracy_dicts = [{
cnn_model_name: accs[i]
} for i, _ in enumerate(self.COMPRESSION_LEVELS)]
self.log_results(
self.pack_eval_values(bits_per_pixel=bpp,
accuracy_dicts=accuracy_dicts))
def _save_results(self, bpp, accs, cnn_model_name):
# write results to csv
csv_file = os.path.join(
self._log_dir,
'{}_{}_{}_accuracy_{}.csv'.format(self._rnn_unit, self._loss_name,
cnn_model_name,
self._log_number))
write_to_csv(csv_file, [(b, *a) for b, a in zip(bpp, accs)])
# log results
accuracy_dicts = [{
cnn_model_name: accs[i]
} for i in range(self._num_iterations)]
self.log_results(
self.pack_eval_values(bits_per_pixel=bpp,
accuracy_dicts=accuracy_dicts))
def _save_results(self, bpp, distortion_dicts, compression_method,
compression_levels):
# write results to csv
csv_file = os.path.join(
self._log_dir, '{}_hvs_{}.csv'.format(compression_method,
self._log_number))
csv_rows = [('bpp', *self.DISTORTION_KEYS)]
csv_rows.extend([(b, *[dist_dict[k] for k in self.DISTORTION_KEYS])
for b, dist_dict in zip(bpp, distortion_dicts)])
write_to_csv(csv_file, csv_rows)
# log results
other_info_dicts = None
if compression_levels is not None:
other_info_dicts = [{'quality': q} for q in compression_levels]
self.log_results(
self.pack_eval_values(bits_per_pixel=bpp,
hvs_dicts=distortion_dicts,
other_info_dicts=other_info_dicts))