Skip to content

Latest commit

 

History

History
86 lines (66 loc) · 3.19 KB

README.md

File metadata and controls

86 lines (66 loc) · 3.19 KB

Benchmark with TNT

Environment Preparation

  1. Create a conda environment and install dependencies as specified in TNT.

  2. Install the apex module.

    git clone https://github.com/ptrblck/apex.git
cd apex
git checkout apex_no_distributed
pip install -v --no-cache-dir ./

  3. Install the Argoverse 1 API.

Data Preparation

  1. Download TFD/TFD-Example into ./dataset/v2x-seq-tfd with ${DATA_ROOT}

    # Download TFD-Example into ./dataset/v2x-seq-tfd
bash tools/dataset_example_download.sh

# export DATA_ROOT, change the DATA_ROOT to TFD-Example data root
export DATA_ROOT=${PWD}'/dataset/v2x-seq-tfd/V2X-Seq-TFD-Example'

  2. Merge Multiple Maps into One Map

    # Merge Multiple Maps into One Map
python tools/data_converter/maps_merge.py --data_root ${DATA_ROOT}

  3. Preprocess Cooperative-view Trajectories

    # Preprocess Cooperative-view Trajectories --split train/val
python tools/trajectory_fusion/fusion_for_prediction.py --data_root ${DATA_ROOT} --split train
python tools/data_converter/tfd_converter.py --data_root ${DATA_ROOT} --split train
python tools/trajectory_fusion/fusion_for_prediction.py --data_root ${DATA_ROOT} --split val
python tools/data_converter/tfd_converter.py --data_root ${DATA_ROOT} --split val

  4. DATA Preprocess

    # Preprocess Raw Data
cd projects/TNT_plugin
python core/util/preprocessor/tfd_preprocess.py --root ${DATA_ROOT}/cooperative-vehicle-infrastructure/fusion_for_prediction

python core/util/preprocessor/tfd_preprocess.py --root ${DATA_ROOT}/cooperative-vehicle-infrastructure/vehicle-trajectories

python core/util/preprocessor/tfd_preprocess.py --root ${DATA_ROOT}/single-vehicle/trajectories

python core/util/preprocessor/tfd_preprocess.py --root ${DATA_ROOT}/single-infrastructure/trajectories

Training

  1. Train the PP-VIC with TNT for Online task

    cd projects/TNT_plugin
bash tnt_train.sh ${GPU_ID} ${DATA_ROOT}/cooperative-vehicle-infrastructure/fusion_for_prediction/interm_data

  2. Train the TNT for Offline task

    cd projects/TNT_plugin

# Pretrained with infrastructure trajectories
bash tnt_pretrain.sh ${GPU_ID} ${DATA_ROOT}/single-infrastructure/trajectories/interm_data

# Finetune with vehicle trajectories and pretrained model
bash tnt_finetune.sh ${GPU_ID} ${DATA_ROOT}/cooperative-vehicle-infrastructure/vehicle-trajectories/interm_data ${PTH}

Evaluation

  1. Evaluate the Trained PP-VIC with TNT for Online task. We have provided the trained online.pth.

    cd projects/TNT_plugin
bash tnt_eval.sh ${GPU_ID} ${DATA_ROOT}/cooperative-vehicle-infrastructure/fusion_for_prediction/interm_data ${PTH}

  2. Evaluate the TNT for Offline task. We have provided the trained offline.pth.

    cd projects/TNT_plugin
bash tnt_eval.sh ${GPU_ID} ${DATA_ROOT}/cooperative-vehicle-infrastructure/vehicle-trajectories/interm_data ${PTH}