gpudrive-main/data_utils/process_nuscenes_files.py

"""Convert NuScenes scenes to GPUDrive-compatible JSON files."""

from __future__ import annotations

import argparse
import logging

try:
    from .nuscenes_adapter import (
        NuScenesAdapterConfig,
        convert_nuscenes_dataset,
    )
except ImportError:  # pragma: no cover - supports running as a script.
    from nuscenes_adapter import NuScenesAdapterConfig, convert_nuscenes_dataset


def _parse_scene_names(value: str | None) -> set[str] | None:
    if not value:
        return None
    return {item.strip() for item in value.split(",") if item.strip()}


def build_arg_parser() -> argparse.ArgumentParser:
    parser = argparse.ArgumentParser(
        description=(
            "Convert NuScenes scenes into the Waymo-style JSON schema consumed "
            "by GPUDrive. Trajectories are interpolated to a fixed 10 Hz, "
            "91-step grid by default."
        )
    )
    parser.add_argument(
        "--dataroot",
        required=True,
        help="Path to the NuScenes dataset root.",
    )
    parser.add_argument(
        "--version",
        default="v1.0-mini",
        help="NuScenes version, e.g. v1.0-mini or v1.0-trainval.",
    )
    parser.add_argument(
        "--output-dir",
        default="data/processed/nuscenes",
        help="Directory where GPUDrive JSON files will be written.",
    )
    parser.add_argument(
        "--split",
        default=None,
        help="Optional NuScenes split name, e.g. mini_train, train, val.",
    )
    parser.add_argument(
        "--scene-names",
        default=None,
        help="Optional comma-separated scene names, e.g. scene-0061,scene-0103.",
    )
    parser.add_argument(
        "--max-scenes",
        type=int,
        default=None,
        help="Optional cap on the number of scenes to convert.",
    )
    parser.add_argument(
        "--file-prefix",
        default="tfrecord-nuscenes",
        help=(
            "Output file prefix. Keep the default if you want SceneDataLoader's "
            "default file_prefix='tfrecord' to discover the files."
        ),
    )
    parser.add_argument(
        "--target-hz",
        type=float,
        default=10.0,
        help="Target interpolation rate in Hz.",
    )
    parser.add_argument(
        "--num-steps",
        type=int,
        default=91,
        help="Number of timesteps per generated GPUDrive scenario.",
    )
    parser.add_argument(
        "--max-annotation-gap-sec",
        type=float,
        default=1.1,
        help="Do not interpolate across raw annotation gaps larger than this.",
    )
    parser.add_argument(
        "--window-stride-sec",
        type=float,
        default=None,
        help=(
            "If set, export multiple 9-second windows per scene using this "
            "stride. If omitted, only the first full window is exported."
        ),
    )
    parser.add_argument(
        "--allow-partial-window",
        action="store_true",
        help="Allow windows that extend beyond the final NuScenes keyframe.",
    )
    parser.add_argument(
        "--no-ego",
        action="store_true",
        help="Do not synthesize an ego vehicle from NuScenes ego poses.",
    )
    parser.add_argument(
        "--no-map",
        action="store_true",
        help="Do not export NuScenes map layers.",
    )
    parser.add_argument(
        "--fail-on-missing-map",
        action="store_true",
        help="Fail conversion if map extraction fails for a scene.",
    )
    parser.add_argument(
        "--map-radius-m",
        type=float,
        default=120.0,
        help="Radius around the scenario center for extracting map features.",
    )
    parser.add_argument(
        "--map-discretization-m",
        type=float,
        default=1.0,
        help="Lane centerline discretization spacing in meters.",
    )
    parser.add_argument(
        "--max-road-elements",
        type=int,
        default=900,
        help="Maximum road records to write per scenario.",
    )
    parser.add_argument(
        "--max-geometry-points",
        type=int,
        default=512,
        help="Maximum points to keep per road geometry.",
    )
    parser.add_argument(
        "--quiet",
        action="store_true",
        help="Reduce NuScenes SDK output.",
    )
    return parser


def main() -> int:
    logging.basicConfig(level=logging.INFO)
    parser = build_arg_parser()
    args = parser.parse_args()

    config = NuScenesAdapterConfig(
        target_hz=args.target_hz,
        num_steps=args.num_steps,
        max_annotation_gap_sec=args.max_annotation_gap_sec,
        require_full_window=not args.allow_partial_window,
        include_ego=not args.no_ego,
        include_map=not args.no_map,
        fail_on_missing_map=args.fail_on_missing_map,
        map_radius_m=args.map_radius_m,
        map_discretization_m=args.map_discretization_m,
        max_road_elements=args.max_road_elements,
        max_geometry_points=args.max_geometry_points,
    )

    stats = convert_nuscenes_dataset(
        dataroot=args.dataroot,
        version=args.version,
        output_dir=args.output_dir,
        config=config,
        split=args.split,
        scene_names=_parse_scene_names(args.scene_names),
        max_scenes=args.max_scenes,
        file_prefix=args.file_prefix,
        window_stride_sec=args.window_stride_sec,
        verbose=not args.quiet,
    )

    logging.info(
        "NuScenes conversion complete: selected=%d, skipped=%d, written=%d",
        stats["scenes_selected"],
        stats["scenes_without_outputs"],
        stats["files_written"],
    )
    return 0


if __name__ == "__main__":
    raise SystemExit(main())