Our code is based on UniAD, we only provide the configuration files related to FusionAD.
Please clone the source code of UniAD into this folder first.
git clone https://github.com/OpenDriveLab/UniAD.git
cd UniAD && git checkout v1.0.1Then follow the instructions for UniAD to install the environment and prepare the dataset.
Once everything is set up, the folder structure should look like this:
The Overall Structure
Please make sure the structure of UniAD is as follows:
FusionAD
├── projects/
├── tools/
├── fusionad_ckpts/
│ ├── epoch_29.pth
├── UniAD/
│ ├── data/
│ │ ├── nuscenes/
│ │ │ ├── can_bus/
│ │ │ ├── maps/
│ │ │ ├── samples/
│ │ │ ├── sweeps/
│ │ │ ├── v1.0-test/
│ │ │ ├── v1.0-trainval/
│ │ ├── infos/
│ │ │ ├── nuscenes_infos_temporal_train.pkl
│ │ │ ├── nuscenes_infos_temporal_val.pkl
│ │ ├── others/
│ │ │ ├── motion_anchor_infos_mode6.pkl
Checkpoint file can be found in here
mkdir ckpts && cd ckpts
wget https://drive.google.com/file/d/1QYPw6L00DKTGJUljOJGzBGAzI0QUZRNa/view?usp=sharingPlease make sure you have prepared the environment and the nuScenes dataset.
cd FusionAD
# single GPU
./tools/dist_eval.sh ./projects/configs/stage2_e2e/fusion_base_e2e.py ./ckpts/fusion_latest.pth 1
# multi GPU
./tools/dist_eval.sh ./projects/configs/stage2_e2e/fusion_base_e2e.py ./ckpts/fusion_latest.pth 8If everything is prepared properly, the output results should be:
NDS: 0.646
mAP: 0.574
amota: 0.502
amotp: 1.059
Motion Metric:
min_ade_err: 0.389
min_fde_err: 0.615
miss_rate_err: 0.084
IoU Metrics:
drivable_iou_mean: 0.732
lanes_iou_mean: 0.368
Occupancy Metrics:
VPQ-n: 64.951 VPQ-f+: 50.330
IoU-n: 70.540 IoU-f+: 50.985
Planning Metrics:
CR_avg: 0.116
L2: 0.708
CR_traj: 0.349
# N_GPUS is the number of GPUs used. Recommended >=8.
./tools/dist_train.sh ./projects/configs/stage2_e2e/fusion_base_e2e.py N_GPUS