search

AIR-THU / DAIR-V2X

Apache License 2.0
424 stars 65 forks source link

Training #75

Closed Ting234 closed 7 months ago

Ting234 commented 10 months ago

Training

Implementation Framework. We directly use MMDetection3D (v0.17.1) to train the infrastructure 3D detector and vehicle 3D detector.

Infrastructure detector training details. Before training the detectors, we should follow MMDetection3D to convert the "./data/DAIR-V2X/cooperative-vehicle-infrastructure/infrastructure-side" into specific training format. Then we train the PointPillars with configure file [trainval_config_i.py](https://github.com/AIR-THU/DAIR-V2X/blob/main/configs/vic3d/late-fusion-pointcloud/pointpillars/trainval_config_i.py)

Vehicle detector training details. Before training the detectors, we should follow MMDetection3D to convert the "./data/DAIR-V2X/cooperative-vehicle-infrastructure/vehicle-side" into specific training format. Then we train the PointPillars with configure file [trainval_config_v.py](https://github.com/AIR-THU/DAIR-V2X/blob/main/configs/vic3d/late-fusion-pointcloud/pointpillars/trainval_config_v.py)

怎么训练呢？该怎么操作呢！求

Ting234 commented 10 months ago

Then we train the PointPillars with configure file [trainval_config_i.py]怎么用呢

haibao-yu commented 7 months ago

Then we train the PointPillars with configure file [trainval_config_i.py]怎么用呢

You can convert the DAIR-V2X dataset into Kitti format as https://github.com/AIR-THU/DAIR-V2X/tree/main/configs/sv3d-inf/pointpillars, and then train and eval the model as mmdetection3d https://mmdetection3d.readthedocs.io/en/latest/user_guides/train_test.html.