forked from PaddlePaddle/Paddle3D
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpointpillars_xyres16_kitti_cyclist_pedestrian.yml
205 lines (199 loc) · 6.09 KB
/
pointpillars_xyres16_kitti_cyclist_pedestrian.yml
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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
batch_size: 2
iters: 296960 # 160 epochs
train_dataset:
type: KittiPCDataset
dataset_root: datasets/KITTI
class_names: [ "Cyclist", "Pedestrian" ]
transforms:
- type: LoadPointCloud
dim: 4
use_dim: 4
- type: RemoveCameraInvisiblePointsKITTI
- type: SamplingDatabase
min_num_points_in_box_per_class:
Cyclist: 5
max_num_samples_per_class:
Cyclist: 8
ignored_difficulty: [ -1 ]
database_anno_path: datasets/KITTI/kitti_train_gt_database/anno_info_train.pkl
database_root: datasets/KITTI/
class_names: [ "Cyclist", "Pedestrian" ]
- type: RandomObjectPerturb
rotation_range: [ -0.15707963267, 0.15707963267 ]
translation_std: [ 0.25, 0.25, 0.25 ]
max_num_attempts: 100
- type: RandomVerticalFlip
- type: GlobalRotate
min_rot: -0.78539816
max_rot: 0.78539816
- type: GlobalScale
min_scale: 0.95
max_scale: 1.05
- type: GlobalTranslate
translation_std: [ 0.2, 0.2, 0.2 ]
- type: FilterBBoxOutsideRange
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
- type: ShufflePoint
- type: HardVoxelize
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
voxel_size: [ 0.16, 0.16, 3 ]
max_points_in_voxel: 100
max_voxel_num: 12000
- type: GenerateAnchors
output_stride_factor: 1 # RPN `downsample_strides`[0] // `upsample_strides`[0]
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
voxel_size: [ 0.16, 0.16, 3 ]
anchor_configs:
- sizes: [ 0.6, 1.76, 1.73 ] # wlh
anchor_strides: [ 0.16, 0.16, 0.0 ]
anchor_offsets: [ 0.08, -19.76, -1.465 ]
rotations: [ 0, 1.57 ]
matched_threshold: 0.5
unmatched_threshold: 0.35
- sizes: [ 0.6, 0.8, 1.73 ] # wlh
anchor_strides: [ 0.16, 0.16, 0.0 ]
anchor_offsets: [ 0.08, -19.76, -1.465 ]
rotations: [ 0, 1.57 ]
matched_threshold: 0.5
unmatched_threshold: 0.35
anchor_area_threshold: 1
- type: Gt2PointPillarsTarget
rpn_batch_size: 512
mode: train
val_dataset:
type: KittiPCDataset
dataset_root: datasets/KITTI
class_names: [ "Cyclist", "Pedestrian" ]
transforms:
- type: LoadPointCloud
dim: 4
use_dim: 4
- type: RemoveCameraInvisiblePointsKITTI
- type: HardVoxelize
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
voxel_size: [ 0.16, 0.16, 3 ]
max_points_in_voxel: 100
max_voxel_num: 12000
- type: GenerateAnchors
output_stride_factor: 1
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
voxel_size: [ 0.16, 0.16, 3 ]
anchor_configs:
- sizes: [ 0.6, 1.76, 1.73 ] # wlh
anchor_strides: [ 0.16, 0.16, 0.0 ]
anchor_offsets: [ 0.08, -19.76, -1.465 ]
rotations: [ 0, 1.57 ]
matched_threshold: 0.5
unmatched_threshold: 0.35
- sizes: [ 0.6, 0.8, 1.73 ] # wlh
anchor_strides: [ 0.16, 0.16, 0.0 ]
anchor_offsets: [ 0.08, -19.76, -1.465 ]
rotations: [ 0, 1.57 ]
matched_threshold: 0.5
unmatched_threshold: 0.35
anchor_area_threshold: 1
mode: val
model:
type: PointPillars
voxelizer:
type: HardVoxelizer
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
voxel_size: [ 0.16, 0.16, 3 ]
max_num_points_in_voxel: 100
max_num_voxels: 12000
pillar_encoder:
type: PillarFeatureNet
in_channels: 4
feat_channels: [ 64 ]
with_distance: False
max_num_points_in_voxel: 100
voxel_size: [ 0.16, 0.16, 3 ]
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
legacy: False
middle_encoder:
type: PointPillarsScatter
in_channels: 64
voxel_size: [ 0.16, 0.16, 3 ]
point_cloud_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
backbone:
type: SecondBackbone
in_channels: 64
out_channels: [ 64, 128, 256 ]
layer_nums: [ 3, 5, 5 ]
downsample_strides: [ 1, 2, 2 ]
neck:
type: SecondFPN
in_channels: [ 64, 128, 256 ]
out_channels: [ 128, 128, 128 ]
upsample_strides: [ 1, 2, 4 ]
use_conv_for_no_stride: False
head:
type: SSDHead
num_classes: 2
feature_channels: 384 # sum(upsample_channels)
num_anchor_per_loc: 4
encode_background_as_zeros: True
use_direction_classifier: True
box_code_size: 7
nms_score_threshold: 0.05
nms_pre_max_size: 1000
nms_post_max_size: 300
nms_iou_threshold: 0.5
prediction_center_limit_range: [ 0, -19.84, -2.5, 47.36, 19.84, 0.5 ]
loss:
type: PointPillarsLoss
num_classes: 2
classification_loss:
type: SigmoidFocalClassificationLoss
gamma: 2.0
alpha: 0.25
regression_loss:
type: WeightedSmoothL1RegressionLoss
sigma: 3.0
code_weights: [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ]
direction_loss:
type: WeightedSoftmaxClassificationLoss
classification_loss_weight: 1.0
regression_loss_weight: 2.0
direction_loss_weight: 0.2
fg_cls_weight: 1.0
bg_cls_weight: 1.0
encode_rot_error_by_sin: True
use_direction_classifier: True
encode_background_as_zeros: True
box_code_size: 7
anchor_configs:
- sizes: [ 0.6, 1.76, 1.73 ] # wlh
anchor_strides: [ 0.16, 0.16, 0.0 ]
anchor_offsets: [ 0.08, -19.76, -1.465 ]
rotations: [ 0, 1.57 ]
matched_threshold: 0.5
unmatched_threshold: 0.35
- sizes: [ 0.6, 0.8, 1.73 ] # wlh
anchor_strides: [ 0.16, 0.16, 0.0 ]
anchor_offsets: [ 0.08, -19.76, -1.465 ]
rotations: [ 0, 1.57 ]
matched_threshold: 0.5
unmatched_threshold: 0.35
anchor_area_threshold: 1
optimizer:
type: Adam
weight_decay: 0.0001
grad_clip:
type: ClipGradByGlobalNorm
clip_norm: 10.0
lr_scheduler:
type: StepDecay
learning_rate: 0.0002
step_size: 27840 # decay every 15 epochs
gamma: 0.8
export:
transforms:
- type: LoadPointCloud
dim: 4
use_dim: 4
- type: HardVoxelize
point_cloud_range: [ 0, -39.68, -3, 69.12, 39.68, 1 ]
voxel_size: [ 0.16, 0.16, 4 ]
max_points_in_voxel: 32
max_voxel_num: 16000