Offline preprocessors for Apairo — ground filtering, odometry, segmentation & derived channels, computed once and persisted

Label synchronized channels — 3D point clouds, camera images — into a top-down BEV grid or per-point labels.

At-access transforms & augmentations for apairo datasets — numpy-in/numpy-out, applied lazily at read time

Head-to-head benchmarks — apairo loads at the same throughput as a hand-rolled loader, for a fraction of the code

Rosbag extractor working with apairo (apairo config, preprocess, ... ) with a CLI interface

Robotics datasets prepared with apairo, published on the Hugging Face Hub — doubling as end-to-end usage examples

Annotate lidar point clouds in 3D — run a model, click one cluster, label the whole group at once.