-
Notifications
You must be signed in to change notification settings - Fork 2.6k
/
upernet_beit-large_fp16_8x1_640x640_160k_ade20k.py
47 lines (42 loc) · 1.3 KB
/
upernet_beit-large_fp16_8x1_640x640_160k_ade20k.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
_base_ = [
'../_base_/models/upernet_beit.py', '../_base_/datasets/ade20k_640x640.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_320k.py'
]
model = dict(
pretrained='pretrain/beit_large_patch16_224_pt22k_ft22k.pth',
backbone=dict(
type='BEiT',
embed_dims=1024,
num_layers=24,
num_heads=16,
mlp_ratio=4,
qv_bias=True,
init_values=1e-6,
drop_path_rate=0.2,
out_indices=[7, 11, 15, 23]),
neck=dict(embed_dim=1024, rescales=[4, 2, 1, 0.5]),
decode_head=dict(
in_channels=[1024, 1024, 1024, 1024], num_classes=150, channels=1024),
auxiliary_head=dict(in_channels=1024, num_classes=150),
test_cfg=dict(mode='slide', crop_size=(640, 640), stride=(426, 426)))
optimizer = dict(
_delete_=True,
type='AdamW',
lr=2e-5,
betas=(0.9, 0.999),
weight_decay=0.05,
constructor='LayerDecayOptimizerConstructor',
paramwise_cfg=dict(num_layers=24, layer_decay_rate=0.95))
lr_config = dict(
_delete_=True,
policy='poly',
warmup='linear',
warmup_iters=3000,
warmup_ratio=1e-6,
power=1.0,
min_lr=0.0,
by_epoch=False)
data = dict(samples_per_gpu=1)
optimizer_config = dict(
type='GradientCumulativeFp16OptimizerHook', cumulative_iters=2)
fp16 = dict()