-
Notifications
You must be signed in to change notification settings - Fork 42
/
mvp_campus.py
221 lines (218 loc) · 6.98 KB
/
mvp_campus.py
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
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
# finetune with the weight pretrained from panoptic dataset
# using cam3, cam12, cam23
__dataset__ = 'campus'
__train_dataset__ = __dataset__
__test_dataset__ = __dataset__
__resnet_n_layer__ = 50
__n_kps__ = 15
__dataset_n_kps__ = 14
__n_instance__ = 10
__n_cameras__ = 3
__image_size__ = [1000, 800]
__space_size__ = [12000.0, 12000.0, 2000.0]
__space_center__ = [3000.0, 4500.0, 1000.0]
__decoder_size__ = 256
__projattn_pos_embed_mode__ = 'use_rayconv'
__pred_conf_threshold__ = 0.5
__n_heads__ = 8
model = 'multi_view_pose_transformer'
output_dir = 'output'
dataset = __dataset__
backbone_layers = __resnet_n_layer__
trainer_setup = dict(
type='MVPTrainer',
workers=4,
train_dataset=__train_dataset__,
test_dataset=__test_dataset__,
lr=0.00003,
lr_linear_proj_mult=0.1,
weight_decay=1e-4,
optimizer='adamw',
end_epoch=25,
pretrained_backbone='',
model_root='./weight/mvp',
finetune_model='xrmocap_mvp_panoptic_3view_3_12_23-4b391740_20220831.pth',
resume=False,
lr_decay_epoch=[10],
inference_conf_thr=[0.0],
train_batch_size=1,
test_batch_size=1,
test_model_file='model_best.pth.tar',
clip_max_norm=0.1,
print_freq=100,
cudnn_setup=dict(
benchmark=True,
deterministic=False,
enable=True,
),
dataset_setup=dict(
train_dataset_setup=dict(
type='MVPDataset',
test_mode=False,
meta_path= # noqa E251
'./xrmocap_data/CampusSeq1/xrmocap_meta_trainset_pesudo_gt',
),
test_dataset_setup=dict(
type='MVPDataset',
test_mode=True,
meta_path='./xrmocap_data/CampusSeq1/xrmocap_meta_testset',
),
base_dataset_setup=dict(
dataset=__dataset__,
data_root='./xrmocap_data/CampusSeq1',
img_pipeline=[
dict(type='LoadImageCV2'),
dict(type='BGR2RGB'),
dict(
type='WarpAffine',
image_size=__image_size__,
flag='inter_linear'),
dict(type='ToTensor'),
dict(
type='Normalize',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
image_size=__image_size__,
heatmap_size=[250, 200],
metric_unit='millimeter',
shuffled=False,
gt_kps3d_convention='campus', # same convention for shelf, campus
cam_world2cam=True,
n_max_person=__n_instance__,
n_views=__n_cameras__,
n_kps=__dataset_n_kps__,
),
),
mvp_setup=dict(
type='MviewPoseTransformer',
n_kps=__n_kps__,
n_instance=__n_instance__,
image_size=__image_size__,
space_size=__space_size__,
space_center=__space_center__,
d_model=__decoder_size__,
use_feat_level=[0, 1, 2],
n_cameras=__n_cameras__,
query_embed_type='person_kp',
with_pose_refine=True,
loss_weight_loss_ce=0.0,
loss_per_kp=5.,
aux_loss=True,
pred_conf_threshold=__pred_conf_threshold__,
pred_class_fuse='mean',
projattn_pos_embed_mode=__projattn_pos_embed_mode__,
query_adaptation=True,
convert_kp_format_indexes=[
14, 13, 12, 6, 7, 8, 11, 10, 9, 3, 4, 5, 0, 1
],
backbone_setup=dict(
type='PoseResNet',
n_layers=__resnet_n_layer__,
n_kps=__n_kps__,
deconv_with_bias=False,
n_deconv_layers=3,
n_deconv_filters=[256, 256, 256],
n_deconv_kernels=[4, 4, 4],
final_conv_kernel=1,
),
proj_attn_setup=dict(
type='ProjAttn',
d_model=__decoder_size__,
n_levels=1,
n_heads=__n_heads__,
n_points=4,
projattn_pos_embed_mode=__projattn_pos_embed_mode__,
),
decoder_layer_setup=dict(
type='MvPDecoderLayer',
space_size=__space_size__,
space_center=__space_center__,
image_size=__image_size__,
d_model=__decoder_size__,
dim_feedforward=1024,
dropout=0.1,
activation='relu',
n_heads=__n_heads__,
detach_refpoints_cameraprj=True,
fuse_view_feats='cat_proj',
n_views=__n_cameras__,
),
decoder_setup=dict(
type='MvPDecoder',
n_decoder_layer=6,
return_intermediate=True,
),
pos_encoding_setup=dict(
type='PositionEmbeddingSine',
normalize=True,
temperature=10000,
),
pose_embed_setup=dict(
type='MLP', d_model=__decoder_size__, pose_embed_layer=3),
matcher_setup=dict(
type='HungarianMatcher',
match_coord='norm',
),
criterion_setup=dict(
type='SetCriterion',
image_size=__image_size__,
n_person=__n_instance__,
loss_kp_type='l1',
focal_alpha=0.25,
space_size=__space_size__,
space_center=__space_center__,
use_loss_pose_perprojection=True,
loss_pose_normalize=False,
pred_conf_threshold=__pred_conf_threshold__,
),
),
evaluation_setup=dict(
type='End2EndEvaluation',
dataset_name=__test_dataset__,
pred_kps3d_convention='campus',
gt_kps3d_convention='campus',
eval_kps3d_convention='campus',
n_max_person=__n_instance__,
checkpoint_select='pcp_total_mean',
metric_list=[
dict(
type='PredictionMatcher',
name='matching',
),
dict(type='MPJPEMetric', name='mpjpe', unit_scale=1),
dict(type='PAMPJPEMetric', name='pa_mpjpe', unit_scale=1),
dict(
type='PCKMetric',
name='pck',
use_pa_mpjpe=True,
threshold=[50, 100],
),
dict(
type='PCPMetric',
name='pcp',
threshold=0.5,
show_table=True,
selected_limbs_names=[
'left_lower_leg', 'right_lower_leg', 'left_upperarm',
'right_upperarm', 'left_forearm', 'right_forearm',
'left_thigh', 'right_thigh'
],
additional_limbs_names=[['jaw', 'headtop']],
),
dict(
type='PrecisionRecallMetric',
name='precision_recall',
show_table=False,
threshold=list(range(25, 155, 25)) + [500],
)
],
pick_dict=dict(
mpjpe=['mpjpe_mean', 'mpjpe_std'],
pa_mpjpe=['pa_mpjpe_mean', 'pa_mpjpe_std'],
pck=['pck@50', 'pck@100'],
pcp=['pcp_total_mean'],
precision_recall=['recall@500'],
),
),
)