File size: 5,509 Bytes
e012b50 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 | _base_ = ['../_base_/default_runtime.py']
custom_imports = dict(
imports=['models.cooperative'],
allow_failed_imports=False)
vis_backends = [dict(type='LocalVisBackend')]
visualizer = dict(
type='Det3DLocalVisualizer', vis_backends=vis_backends, name='visualizer')
voxel_size = [0.4, 0.4, 4]
point_cloud_range = [-140.8, -40, -3, 140.8, 40, 1]
gt_range = [-140, -40, -10, 140, 40, 10]
model_args = dict(
max_cav=5,
lidar_range=point_cloud_range,
voxel_size=voxel_size,
anchor_number=2,
backbone_fix=False,
compression=0,
pillar_vfe=dict(
num_filters=[64],
use_absolute_xyz=True,
use_norm=True,
with_distance=False),
point_pillar_scatter=dict(
num_features=64,
grid_size=[704, 200, 1]),
base_bev_backbone=dict(
layer_nums=[3, 5, 8],
layer_strides=[2, 2, 2],
num_filters=[64, 128, 256],
num_upsample_filter=[128, 128, 128],
upsample_strides=[1, 2, 4]),
shrink_header=dict(
kernal_size=[3],
stride=[2],
padding=[1],
dim=[256],
input_dim=384),
v2vfusion=dict(
agg_operator='avg',
conv_gru=dict(H=50, W=176, kernel_size=[[3, 3]], num_layers=1),
downsample_rate=4,
gru_flag=True,
in_channels=256,
num_iteration=2,
voxel_size=voxel_size,
),
)
anchor_args = dict(
D=1, H=200, W=704,
l=3.9, w=1.6, h=1.56,
num=2, r=[0, 90],
cav_lidar_range=point_cloud_range,
feature_stride=4,
vd=4, vh=0.4, vw=0.4,
)
postprocess_args = dict(
max_num=100, nms_thresh=0.15,
target_args=dict(
pos_threshold=0.6, neg_threshold=0.45, score_threshold=0.20),
)
loss_args = dict(cls_weight=1.0, reg=2.0)
model = dict(
type='CooperativeDetector',
arch='v2vnet',
max_cav=5,
model_args=model_args,
anchor_args=anchor_args,
postprocess_args=postprocess_args,
loss_args=loss_args,
data_preprocessor=dict(
type='SpVoxelCoopDet3DDataPreprocessor',
voxel=True,
voxel_layer=dict(
max_num_points=32,
point_cloud_range=point_cloud_range,
voxel_size=voxel_size,
max_voxels=(32000, 70000)),
cav_lidar_range=point_cloud_range,
voxel_size=voxel_size,
max_points_per_voxel=32,
max_voxel_train=32000,
max_voxel_test=70000),
bbox_head=dict(
type='DetHead',
in_channels=256,
anchor_number=2,
anchor_size=[3.9, 1.6, 1.56],
anchor_rotations=[0, 90],
anchor_z=-1.0,
point_cloud_range=point_cloud_range,
voxel_size=voxel_size,
feature_stride=4,
pos_threshold=0.6,
neg_threshold=0.45,
score_threshold=0.20,
nms_threshold=0.15,
max_num=100,
cls_weight=1.0,
reg_weight=2.0),
train_cfg=None,
test_cfg=None)
dataset_type = 'CoopDataset'
data_root = 'data/cooperscene'
train_pipeline = [
dict(type='LoadCooperativePointCloud', coord_type='LIDAR',
load_dim=4, use_dim=[0, 1, 2, 3], max_cav=5,
proj_first=True,
point_cloud_range=point_cloud_range),
dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
dict(type='PackCooperative3DDetInputs',
keys=['gt_bboxes_3d', 'gt_labels_3d']),
]
test_pipeline = [
dict(type='LoadCooperativePointCloud', coord_type='LIDAR',
load_dim=4, use_dim=[0, 1, 2, 3], max_cav=5,
proj_first=True,
point_cloud_range=point_cloud_range),
dict(type='PackCooperative3DDetInputs', keys=[]),
]
train_dataloader = dict(
batch_size=4, collate_fn=dict(type='cooperative_collate'),
num_workers=4, sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type, data_root=data_root,
ann_file='cooperscene_coop_infos_train.pkl',
data_prefix=dict(pts=''), pipeline=train_pipeline,
pcd_limit_range=point_cloud_range, max_cav=5, com_range=70))
val_dataloader = dict(
batch_size=4, collate_fn=dict(type='cooperative_collate'),
num_workers=4, sampler=dict(type='DefaultSampler', shuffle=False),
dataset=dict(
type=dataset_type, data_root=data_root,
ann_file='cooperscene_coop_infos_val.pkl',
data_prefix=dict(pts=''), pipeline=test_pipeline,
test_mode=True, pcd_limit_range=gt_range, max_cav=5, com_range=70))
test_dataloader = dict(
batch_size=4, collate_fn=dict(type='cooperative_collate'),
num_workers=4, sampler=dict(type='DefaultSampler', shuffle=False),
dataset=dict(
type=dataset_type, data_root=data_root,
ann_file='cooperscene_coop_infos_test.pkl',
data_prefix=dict(pts=''), pipeline=test_pipeline,
test_mode=True, pcd_limit_range=gt_range, max_cav=5, com_range=70))
val_evaluator = dict(type='EvalMetric')
test_evaluator = dict(type='EvalMetric')
load_from = None
optim_wrapper = dict(
type='OptimWrapper',
optimizer=dict(type='Adam', lr=1e-4, eps=1e-10, weight_decay=1e-4))
param_scheduler = [
dict(type='LinearLR', start_factor=0.2, by_epoch=True, begin=0, end=3),
dict(type='CosineAnnealingLR', by_epoch=True, begin=3, end=30,
eta_min=1e-6),
]
train_cfg = dict(by_epoch=True, max_epochs=30, val_interval=1)
val_cfg = dict()
test_cfg = dict()
default_hooks = dict(checkpoint=dict(type='CheckpointHook', interval=1))
|