bev-project/configs/nuscenes/three_tasks/default.yaml

# 三任务基础配置
# 任务: 3D检测 + BEV分割 + 矢量地图

_base_:
  - ../../det/transfusion/secfpn/camera+lidar/default.yaml

# 全局参数
voxel_size: [0.075, 0.075, 0.2]
point_cloud_range: [-54.0, -54.0, -5.0, 54.0, 54.0, 3.0]
bev_h: 180
bev_w: 180

# 图像参数
image_size: [256, 704]

# BEV范围（用于矢量地图）
bev_x_range: [-50, 50]
bev_y_range: [-50, 50]

# 矢量地图参数
num_vec_classes: 3  # divider, boundary, ped_crossing
max_num_vectors: 50
num_points_per_vec: 20

# nuScenes类别
object_classes:
  - car
  - truck
  - construction_vehicle
  - bus
  - trailer
  - barrier
  - motorcycle
  - bicycle
  - pedestrian
  - traffic_cone

map_classes:
  - drivable_area
  - ped_crossing
  - walkway
  - stop_line
  - carpark_area
  - divider

# 数据增强参数
augment2d:
  resize: [[0.38, 0.55], [0.48, 0.48]]
  rotate: [-5.4, 5.4]
  gridmask:
    prob: 0.0
    fixed_prob: true

augment3d:
  scale: [0.9, 1.1]
  rotate: [-0.78539816, 0.78539816]
  translate: 0.5

# 优化器参数
optimizer:
  type: AdamW
  lr: 2.0e-4
  weight_decay: 0.01

optimizer_config:
  grad_clip:
    max_norm: 35
    norm_type: 2

lr_config:
  policy: cyclic
  target_ratio: 5.0
  cyclic_times: 1
  step_ratio_up: 0.4

momentum_config:
  policy: cyclic
  target_ratio: 0.8 / 0.95
  cyclic_times: 1
  step_ratio_up: 0.4

# 训练参数
max_epochs: 20
batch_size_per_gpu: 1
num_workers_per_gpu: 4
checkpoint_interval: 1

# 数据集
data:
  samples_per_gpu: ${batch_size_per_gpu}
  workers_per_gpu: ${num_workers_per_gpu}
  
  train:
    type: NuScenesDataset
    dataset_root: data/nuscenes
    ann_file: data/nuscenes/nuscenes_infos_train.pkl
    pipeline: ${train_pipeline}
    object_classes: ${object_classes}
    map_classes: ${map_classes}
    modality:
      use_camera: true
      use_lidar: true
      use_radar: false
      use_map: false
      use_external: false
  
  val:
    type: NuScenesDataset
    dataset_root: data/nuscenes
    ann_file: data/nuscenes/nuscenes_infos_val.pkl
    pipeline: ${val_pipeline}
    object_classes: ${object_classes}
    map_classes: ${map_classes}
    modality:
      use_camera: true
      use_lidar: true
      use_radar: false
      use_map: false
      use_external: false

# 训练pipeline
train_pipeline:
  - type: LoadMultiViewImageFromFiles
    to_float32: true
  - type: LoadPointsFromFile
    coord_type: LIDAR
    load_dim: ${load_dim}
    use_dim: ${use_dim}
  - type: LoadPointsFromMultiSweeps
    sweeps_num: 9
    use_dim: [0, 1, 2, 3, 4]
    pad_empty_sweeps: true
    remove_close: true
  - type: LoadAnnotations3D
    with_bbox_3d: true
    with_label_3d: true
    with_attr_label: false
  - type: LoadBEVSegmentation
    dataset_root: data/nuscenes
    xbound: [-50.0, 50.0, 0.5]
    ybound: [-50.0, 50.0, 0.5]
    classes: ${map_classes}
  - type: LoadVectorMap  # 🆕 矢量地图加载
    vector_map_file: data/nuscenes/vector_maps_bevfusion.pkl
    num_vec_classes: ${num_vec_classes}
    max_num_vectors: ${max_num_vectors}
    num_points_per_vec: ${num_points_per_vec}
  - type: ObjectRangeFilter
    point_cloud_range: ${point_cloud_range}
  - type: ObjectNameFilter
    classes: ${object_classes}
  - type: ImageAug3D
    final_dim: ${image_size}
    resize_lim: ${augment2d.resize[0]}
    bot_pct_lim: [0.0, 0.0]
    rot_lim: ${augment2d.rotate}
    rand_flip: true
    is_train: true
  - type: GlobalRotScaleTrans
    resize_lim: ${augment3d.scale}
    rot_lim: ${augment3d.rotate}
    trans_lim: ${augment3d.translate}
    is_train: true
  - type: PointsRangeFilter
    point_cloud_range: ${point_cloud_range}
  - type: PointShuffle
  - type: DefaultFormatBundle3D
    classes: ${object_classes}
  - type: Collect3D
    keys:
      - img
      - points
      - gt_bboxes_3d
      - gt_labels_3d
      - gt_masks_bev
      - gt_vectors_labels   # 🆕
      - gt_vectors_points   # 🆕
    meta_keys:
      - camera_intrinsics
      - camera2ego
      - lidar2ego
      - lidar2camera
      - camera2lidar
      - lidar2image
      - img_aug_matrix
      - lidar_aug_matrix

# 验证pipeline
val_pipeline:
  - type: LoadMultiViewImageFromFiles
    to_float32: true
  - type: LoadPointsFromFile
    coord_type: LIDAR
    load_dim: ${load_dim}
    use_dim: ${use_dim}
  - type: LoadPointsFromMultiSweeps
    sweeps_num: 9
    use_dim: [0, 1, 2, 3, 4]
    pad_empty_sweeps: true
    remove_close: true
  - type: LoadAnnotations3D
    with_bbox_3d: true
    with_label_3d: true
    with_attr_label: false
  - type: LoadBEVSegmentation
    dataset_root: data/nuscenes
    xbound: [-50.0, 50.0, 0.5]
    ybound: [-50.0, 50.0, 0.5]
    classes: ${map_classes}
  - type: LoadVectorMap  # 🆕
    vector_map_file: data/nuscenes/vector_maps_bevfusion.pkl
    num_vec_classes: ${num_vec_classes}
    max_num_vectors: ${max_num_vectors}
    num_points_per_vec: ${num_points_per_vec}
  - type: ImageAug3D
    final_dim: ${image_size}
    resize_lim: ${augment2d.resize[1]}
    bot_pct_lim: [0.0, 0.0]
    rot_lim: [0.0, 0.0]
    rand_flip: false
    is_train: false
  - type: GlobalRotScaleTrans
    resize_lim: [1.0, 1.0]
    rot_lim: [0.0, 0.0]
    trans_lim: 0.0
    is_train: false
  - type: PointsRangeFilter
    point_cloud_range: ${point_cloud_range}
  - type: DefaultFormatBundle3D
    classes: ${object_classes}
  - type: Collect3D
    keys:
      - img
      - points
      - gt_bboxes_3d
      - gt_labels_3d
      - gt_masks_bev
      - gt_vectors_labels   # 🆕
      - gt_vectors_points   # 🆕
    meta_keys:
      - camera_intrinsics
      - camera2ego
      - lidar2ego
      - lidar2camera
      - camera2lidar
      - lidar2image
      - img_aug_matrix
      - lidar_aug_matrix