Instructions for FSDv2

Resources

We provide complete first-hand resources on all three datasets to reproduce our performance and track the training process, including checkpoints, logs, results.

Access the resources at: https://share.weiyun.com/1HgbFpyI or https://drive.google.com/drive/folders/17xG_AVqCOTzPPKl6RNHQyXlwG8hmCyJC?usp=sharing

If the link expires, feel free to open an issue. Due to the Waymo license, please contact Lue Fan (fanlue2019@ia.ac) to access the Waymo resources privately.

---

Data Preparation

Follow official MMDetection3Dv0.15.0 to prepare data: https://github.com/open-mmlab/mmdetection3d/releases/tag/v0.15.0 . Note that users do not need to install the official MMDetection3D, just following their instructions to prepare data. If you have used this SST repo before, please skip this step.

Run Experiments

Users only need to change the content in run.sh to:

DIR=fsdv2
WORK=work_dirs

# for waymo
CONFIG=fsdv2_waymo_2x
bash tools/dist_train.sh configs/$DIR/$CONFIG.py 8 --work-dir ./$WORK/$CONFIG/ --cfg-options evaluation.pklfile_prefix=./$WORK/$CONFIG/results evaluation.metric=fast --seed 1

# for argoverse 2
CONFIG=fsdv2_argo_2x
bash tools/dist_train.sh configs/$DIR/$CONFIG.py 8 --work-dir ./$WORK/$CONFIG/ --cfg-options evaluation.pklfile_prefix=./$WORK/$CONFIG/results evaluation.metric=fast --seed 1

# for nuscenes
CONFIG=fsdv2_nusc_2x
bash tools/dist_train.sh configs/$DIR/$CONFIG.py 8 --work-dir ./$WORK/$CONFIG/ --cfg-options evaluation.jsonfile_prefix=./$WORK/$CONFIG/results evaluation.metric=bbox --seed 1