search

Owen-Liuyuxuan / visualDet3D

Official Repo for Ground-aware Monocular 3D Object Detection for Autonomous Driving / YOLOStereo3D: A Step Back to 2D for Efficient Stereo 3D Detection
https://owen-liuyuxuan.github.io/papers_reading_sharing.github.io/3dDetection/GroundAwareConvultion/
Apache License 2.0
365 stars 77 forks source link

How can i get test rusults #14

Closed Yongsz closed 3 years ago

Yongsz commented 3 years ago

How can I get my prediction results on the test data? I try to run the testdata with eval.sh but it shows that there is no label. How can I solve it?

Owen-Liuyuxuan commented 3 years ago

You need to submit to the test server of KITTI. First get an account in http://www.cvlibs.net/datasets/kitti/user_submit.php Then "cd" to "workdirs/$name/output/test/data".

cd workdirs/$name/output/test/data

zip data.zip *.txt

you can upload the data.zip to the KITTI website to get your result.

Yongsz commented 3 years ago

how can i get the test data?Thanks!

Owen-Liuyuxuan commented 3 years ago

  1. You may need to convert the test-set data to imdb that we are using in the current repo. You need to run 

    ./launcher/det_precompute.sh config/$CONFIG_FILE.py test # only for upload testing

    as mentions in mono3d

  2. The test label of KITTI dataset is not public, we only get results from the server, but we can not obtain ground truth. That is just for fair competition. As a result, as coded in this repo, we only produce the prediction results on test data and we can not generate scores locally. We need to submit the produced data to the KITTI server to get a score remotely.

  3. When you launch 

    ./launcher/eval.sh --config/$CONFIG_FILE.py 0 $CHECKPOINT_PATH validation

    we will validate part of the "training" data of the official split, the split is indicated in the split files, based on your config file.