/
EDTER_BIMLA_320x320_80k_bsds_aug_local8x8_bs_8.py
109 lines (107 loc) · 3.28 KB
/
EDTER_BIMLA_320x320_80k_bsds_aug_local8x8_bs_8.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
_base_ = [
'../_base_/models/edter_bimla_local8x8.py',
'../_base_/datasets/bsds.py', '../_base_/default_runtime.py',
'../_base_/schedules/schedule_80k.py'
]
model = dict(
backbone=dict(img_size=160, pos_embed_interp=True, drop_rate=0., mla_channels=256, mla_index=(2, 5, 8, 11)),
decode_head=dict(img_size=160,mla_channels=256,mlahead_channels=128,num_classes=1),
auxiliary_head=[
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=0,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True, loss_weight=0.4)),
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=1,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True, loss_weight=0.4)),
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=2,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True, loss_weight=0.4)),
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=3,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True , loss_weight=0.4)),
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=4,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True, loss_weight=0.4)),
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=5,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True, loss_weight=0.4)),
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=6,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True, loss_weight=0.4)),
dict(
type='VIT_BIMLA_AUXIHead_LOCAL8x8',
in_channels=256,
channels=512,
in_index=7,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True , loss_weight=0.4)),
],
fuse_head=dict(
type='Local8x8_fuse_head',
in_channels=128,
channels=128,
img_size=160,
num_classes=1,
align_corners=False,
loss_decode=dict(
type='HEDLoss', use_sigmoid=True , loss_weight=1.0))
)
optimizer = dict(lr=1e-6, weight_decay=0.0002,
paramwise_cfg = dict(custom_keys={'head': dict(lr_mult=10.),'global_model': dict(lr_mult=0.),})
)
lr_config = dict(policy='poly', power=0.9, min_lr=1e-8, by_epoch=False)
test_cfg = dict(mode='slide', crop_size=(160, 160), stride=(160, 160))
find_unused_parameters = True
data = dict(samples_per_gpu=4)