This repository contains the official implementation of the paper

UniFlow: Towards Zero-Shot LiDAR Scene Flow for Autonomous Vehicles via Cross-Domain Generalization [arXiv] · [Project Page]

Please follow the environment setup and data preparation instructions provided in OpenSceneFlow.

Beam ID is required for beam-dropout augmentation during training. It is supported for:

• Argoverse 2
• Waymo

Beam ID is not supported for TruckScenes due to information availability, and nuScenes as it is already a sparse 32-beam LiDAR dataset.

If you have not preprocessed Argoverse 2, please run:

python dataprocess/extract_av2.py \
  --av2_type sensor \
  --data_mode train \
  --argo_dir /path/to/av2 \
  --output_dir /path/to/av2/preprocess

If you already have preprocessed Argoverse 2 '.h5' files without beam information, add '--only_insert_beamid' to your command.

If you have not preprocessed Waymo, please run:

python dataprocess/extract_waymo.py \
  --mode train \
  --flow_data_dir /path/to/waymo/flowlabel \
  --map_dir /path/to/waymo/flowlabel/map \
  --output_dir /path/to/waymo/preprocess \
  --nproc <N>

If you already have preprocessed Waymo '.h5' files without beam information, you can insert beam IDs post-hoc using:

python dataprocess/add_beam_id_waymo.py \
  --preprocess_dir /path/to/waymo/preprocess/train \
  --tfrecord_root /path/to/waymo/flowlabel \
  --split train \
  --num_workers <N>

Joint training across multiple datasets is supported via train.py, with model backbone size and data augmentations controlled through configuration options.

Example:

python train.py \
  model=flow4d \
  train_data="[
    /scratch/siyili/waymo/preprocess/train/,
    /scratch/siyili/av2/preprocess/train/,
    /scratch/siyili/nuscenes/preprocess/train/
  ]" \
  val_data=/scratch/siyili/av2/preprocess/val/ \
  optimizer.lr=5e-4 \
  epochs=18 \
  batch_size=2 \
  num_frames=5 \
  model.target.backbone=xl \
  basic_augment=true \
  beam_dropout.enable=true \
  dataset_weight_mode=proportional

If you find UniFlow useful in your research, please consider citing:

@misc{li2025uniflowzeroshotlidarscene,
      title={UniFlow: Towards Zero-Shot LiDAR Scene Flow for Autonomous Vehicles via Cross-Domain Generalization}, 
      author={Siyi Li and Qingwen Zhang and Ishan Khatri and Kyle Vedder and Deva Ramanan and Neehar Peri},
      year={2025},
      eprint={2511.18254},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.18254}, 
}

This repository is based on OpenSceneFlow and includes code adapted from Flow4D and DeFlow. Original copyright and license notices are preserved in the source files where applicable.