This project is forked from https://github.com/wayveai/fiery.
-
git clone this repo and create the necessary directorys
https://github.com/sty61010/fiery.git cd fiery/ mkdir -p data tensorboard_logs
-
Download NuScenes dataset and place it under the
data
directory. -
Create the python environment from
bev18.yml
conda env create -n fiery --file bev18.yml
-
Install
mmdetection3d
package. For more information, please refer to mmdet3d document.git clone https://github.com/open-mmlab/mmdetection3d.git cd mmdetection3d pip install -v -e .
Download the pretrained weight here and extract them into tensorboard_logs
.
In the fiery
directory, run
bash test.sh tensorboard_logs/model 20.ckpt
It will run evaluation on the trainval dataset and report the mAP score.
Expected result:
There are many configs in the fiery/config
folder, you can train the model by
python train.py --config CONFIG GPUS "[0]" ...[Other Parameters]
To preproduce our pretrained model, run
python train.py --config tensorboard_logs/model/hparams.yml