def eval(self,
exp_dir,
predictions,
runtimes,
label=None,
only_metrics=False):
pred_filename = 'jst_predictions_{}.csv'.format(label)
self.save_tusimple_predictions(predictions, runtimes, pred_filename)
if self.metric == 'default':
result = json.loads(
LaneEval.bench_one_submit(pred_filename, self.anno_files[0]))
elif self.metric == 'ours':
result = json.loads(
eval_json(pred_filename,
self.anno_files[0],
json_type='tusimple'))
table = {}
for metric in result:
table[metric['name']] = [metric['value']]
table = tabulate(table, headers='keys')
if not only_metrics:
filename = 'tusimple_{}_eval_result_{}.json'.format(
self.split, label)
with open(os.path.join(exp_dir, filename), 'w') as out_file:
json.dump(result, out_file)
return table, result
def get_metrics(self, lanes, idx):
label = self.annotations[idx]
org_anno = label['old_anno']
pred = self.pred2lanes(org_anno['path'], lanes, org_anno['y_samples'])
_, _, _, matches, accs, dist = LaneEval.bench(pred, org_anno['org_lanes'], org_anno['y_samples'], 0, True)
return matches, accs, dist
# Print the durations
print('The forward pass time for one image is: {}ms'.format(forward_avg *
1000))
print('The clustering time for one image is: {}ms'.format(cluster_avg * 1000))
print('The total time for one image is: {}ms'.format(
(cluster_avg + forward_avg) * 1000))
print('The speed for forward pass is: {}fps'.format(1 / forward_avg))
print('The speed for forward pass is: {}fps'.format(1 / cluster_avg))
# Write the results in the pred.json file
# or in the pred_aug.json file in case you used
# the model trained on augmented data
# Uncomment corresponding line in second case
with open('TUSIMPLE/pred.json', 'w') as f:
#with open('TUSIMPLE/pred_aug.json', 'w') as f:
for res in json_pred:
json.dump(res, f)
f.write('\n')
# Evaluate our results using the TUSIMPLE competition script
# and their groundtruths
from utils.lane import LaneEval
# Uncomment corresponding line in case
# you used the model trained on augmented data
result = LaneEval.bench_one_submit('TUSIMPLE/pred.json',
'TUSIMPLE/test_set/test_label.json')
#result = LaneEval.bench_one_submit('TUSIMPLE/pred_aug.json','TUSIMPLE/test_set/test_label.json')
print(result)