benchmark-HDMapping-AILoopClosure-TransLO

Reproducible setup for TransLO: A Window-Based Masked Point Transformer Framework for Large-Scale LiDAR Odometry.

Only inference.

Requirements

• Ubuntu 24.04
• NVIDIA GPU and CUDA Toolkit 12.x
• Kitti Velodyne dataset (download)
• Docker (follow install instructions)
• Docker NVIDIA runtime (follow install instructions)
• Pretrained model

Running

# clone repository
git clone https://github.com/MapsHD/benchmark-HDMapping-AILoopClosure-TransLO.git

# enter repository
cd benchmark-HDMapping-AILoopClosure-TransLO

# build docker image
docker build -t translo .

# run docker image
docker run --rm -it --gpus all -v PATH_TO_KITTY_DATASET:/DATASET_DIR:ro -v PATH_TO_MODEL_DIRECTORY:/CHECKPOINT_DIR:ro -v "$(pwd)":/OUTPUT_DIR translo

# inside docker container
python3 infer_seq_abs.py --checkpoint /CHECKPOINT_DIR/translo_model_047_-20.478760.pth.tar --sequence /DATASET_DIR/data_odometry_velodyne/dataset/sequences/00/velodyne/ --output /OUTPUT_DIR/output.txt --no-relative