def save_results(args, dataset_name, eval_exp_path, dataset_results, print_header=True):
all_ious, elapsed_time = dataset_results
mean_spc, mean_spi = utils.get_time_metrics(all_ious, elapsed_time)
iou_thrs = np.arange(0.8, min(0.95, args.target_iou) + 0.001, 0.05).tolist()
noc_list, over_max_list = utils.compute_noc_metric(all_ious, iou_thrs=iou_thrs, max_clicks=args.n_clicks)
header, table_row = utils.get_results_table(noc_list, over_max_list, args.mode, dataset_name,
mean_spc, elapsed_time, args.n_clicks,
model_name=eval_exp_path.stem)
target_iou_int = int(args.target_iou * 100)
if target_iou_int not in [80, 85, 90]:
noc_list, over_max_list = utils.compute_noc_metric(all_ious, iou_thrs=[args.target_iou],
max_clicks=args.n_clicks)
table_row += f' NoC@{args.target_iou:.1%} = {noc_list[0]:.2f};'
table_row += f' >={args.n_clicks}@{args.target_iou:.1%} = {over_max_list[0]}'
if print_header:
print(header)
print(table_row)
log_path = eval_exp_path / f'results_{args.mode}_{args.n_clicks}.txt'
if log_path.exists():
with open(log_path, 'a') as f:
f.write(table_row + '\n')
else:
with open(log_path, 'w') as f:
f.write(header + '\n')
f.write(table_row + '\n')
ious_path = eval_exp_path / 'all_ious'
ious_path.mkdir(exist_ok=True)
with open(ious_path / f'{dataset_name}_{args.mode}_{args.n_clicks}.pkl', 'wb') as fp:
pickle.dump(all_ious, fp)
def save_results(args,
row_name,
dataset_name,
logs_path,
logs_prefix,
dataset_results,
save_ious=False,
print_header=True,
single_model_eval=False):
all_ious, elapsed_time = dataset_results
mean_spc, mean_spi = utils.get_time_metrics(all_ious, elapsed_time)
iou_thrs = np.arange(0.8,
min(0.95, args.target_iou) + 0.001, 0.05).tolist()
noc_list, over_max_list = utils.compute_noc_metric(
all_ious, iou_thrs=iou_thrs, max_clicks=args.n_clicks)
row_name = 'last' if row_name == 'last_checkpoint' else row_name
model_name = str(logs_path.relative_to(
args.logs_path)) + ':' + logs_prefix if logs_prefix else logs_path.stem
header, table_row = utils.get_results_table(noc_list,
over_max_list,
row_name,
dataset_name,
mean_spc,
elapsed_time,
args.n_clicks,
model_name=model_name)
if args.print_ious:
min_num_clicks = min(len(x) for x in all_ious)
mean_ious = np.array([x[:min_num_clicks]
for x in all_ious]).mean(axis=0)
miou_str = ' '.join([
f'mIoU@{click_id}={mean_ious[click_id - 1]:.2%};'
for click_id in [1, 2, 3, 5, 10, 20] if click_id <= min_num_clicks
])
table_row += '; ' + miou_str
else:
target_iou_int = int(args.target_iou * 100)
if target_iou_int not in [80, 85, 90]:
noc_list, over_max_list = utils.compute_noc_metric(
all_ious, iou_thrs=[args.target_iou], max_clicks=args.n_clicks)
table_row += f' NoC@{args.target_iou:.1%} = {noc_list[0]:.2f};'
table_row += f' >={args.n_clicks}@{args.target_iou:.1%} = {over_max_list[0]}'
if print_header:
print(header)
print(table_row)
if save_ious:
ious_path = logs_path / 'ious' / (logs_prefix if logs_prefix else '')
ious_path.mkdir(parents=True, exist_ok=True)
with open(
ious_path /
f'{dataset_name}_{args.eval_mode}_{args.mode}_{args.n_clicks}.pkl',
'wb') as fp:
pickle.dump(all_ious, fp)
name_prefix = ''
if logs_prefix:
name_prefix = logs_prefix + '_'
if not single_model_eval:
name_prefix += f'{dataset_name}_'
log_path = logs_path / f'{name_prefix}{args.eval_mode}_{args.mode}_{args.n_clicks}.txt'
if log_path.exists():
with open(log_path, 'a') as f:
f.write(table_row + '\n')
else:
with open(log_path, 'w') as f:
if print_header:
f.write(header + '\n')
f.write(table_row + '\n')