We want to examine the correctness of our segmentation gt of validation set. An extreme segmentation performance fallback on test set is observed, while the detection acts well. Thus we need to confirm that our segmentation gt is correct, although we decoded them following the bevfusion pipeline.
Posted by: terrywayne @ Sept. 21, 2022, 3:29 a.m.Codalab can only support one phase of the competition at one time, so when test phase starts, val phase closed automatically.
If you want to valid your validation performance, you can generate val_gt by bevfusion code based on our validation data and validate your performance using code released on https://github.com/SSLAD2021/AutoScenes-eval, it is the same with our evaluation server.