mirrors/mmsegmentation
1
0
Fork 0
mirror of https://github.com/open-mmlab/mmsegmentation.git synced 2025-06-03 22:03:48 +08:00
Code Issues Projects Releases Wiki Activity

Merge branch 'limengzhang/fix_optim_wrapper' into 'refactor_dev'

Browse Source 
[Fix] Remove OptimizerHook and Add it in OptimWrapper

See merge request openmmlab-enterprise/openmmlab-ce/mmsegmentation!47
This commit is contained in:
zhengmiao 2022-06-16 13:24:19 +00:00
commit 46723a9543
24 changed files with 97 additions and 120 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
configs/_base_/schedules/schedule_160k.py
View File

@ -1,6 +1,6 @@
# optimizer # optimizer
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005) optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer, clip_grad=None)
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(
@ -17,7 +17,6 @@ train_cfg = dict(
val_cfg = dict(type='ValLoop') val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop') test_cfg = dict(type='TestLoop')
default_hooks = dict( default_hooks = dict(
optimizer=dict(type='OptimizerHook', grad_clip=None),
timer=dict(type='IterTimerHook'), timer=dict(type='IterTimerHook'),
logger=dict(type='LoggerHook', interval=50), logger=dict(type='LoggerHook', interval=50),
param_scheduler=dict(type='ParamSchedulerHook'), param_scheduler=dict(type='ParamSchedulerHook'),

3
configs/_base_/schedules/schedule_20k.py
View File

@ -1,6 +1,6 @@
# optimizer # optimizer
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005) optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer, clip_grad=None)
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(
@ -16,7 +16,6 @@ train_cfg = dict(type='IterBasedTrainLoop', max_iters=20000, val_interval=2000)
val_cfg = dict(type='ValLoop') val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop') test_cfg = dict(type='TestLoop')
default_hooks = dict( default_hooks = dict(
optimizer=dict(type='OptimizerHook', grad_clip=None),
timer=dict(type='IterTimerHook'), timer=dict(type='IterTimerHook'),
logger=dict(type='LoggerHook', interval=50), logger=dict(type='LoggerHook', interval=50),
param_scheduler=dict(type='ParamSchedulerHook'), param_scheduler=dict(type='ParamSchedulerHook'),

3
configs/_base_/schedules/schedule_320k.py
View File

@ -1,6 +1,6 @@
# optimizer # optimizer
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005) optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer, clip_grad=None)
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(
@ -17,7 +17,6 @@ train_cfg = dict(
val_cfg = dict(type='ValLoop') val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop') test_cfg = dict(type='TestLoop')
default_hooks = dict( default_hooks = dict(
optimizer=dict(type='OptimizerHook', grad_clip=None),
timer=dict(type='IterTimerHook'), timer=dict(type='IterTimerHook'),
logger=dict(type='LoggerHook', interval=50), logger=dict(type='LoggerHook', interval=50),
param_scheduler=dict(type='ParamSchedulerHook'), param_scheduler=dict(type='ParamSchedulerHook'),

3
configs/_base_/schedules/schedule_40k.py
View File

@ -1,6 +1,6 @@
# optimizer # optimizer
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005) optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer, clip_grad=None)
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(
@ -16,7 +16,6 @@ train_cfg = dict(type='IterBasedTrainLoop', max_iters=40000, val_interval=4000)
val_cfg = dict(type='ValLoop') val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop') test_cfg = dict(type='TestLoop')
default_hooks = dict( default_hooks = dict(
optimizer=dict(type='OptimizerHook', grad_clip=None),
timer=dict(type='IterTimerHook'), timer=dict(type='IterTimerHook'),
logger=dict(type='LoggerHook', interval=50), logger=dict(type='LoggerHook', interval=50),
param_scheduler=dict(type='ParamSchedulerHook'), param_scheduler=dict(type='ParamSchedulerHook'),

3
configs/_base_/schedules/schedule_80k.py
View File

@ -1,6 +1,6 @@
# optimizer # optimizer
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005) optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer, clip_grad=None)
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(
@ -16,7 +16,6 @@ train_cfg = dict(type='IterBasedTrainLoop', max_iters=80000, val_interval=8000)
val_cfg = dict(type='ValLoop') val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop') test_cfg = dict(type='TestLoop')
default_hooks = dict( default_hooks = dict(
optimizer=dict(type='OptimizerHook', grad_clip=None),
timer=dict(type='IterTimerHook'), timer=dict(type='IterTimerHook'),
logger=dict(type='LoggerHook', interval=50), logger=dict(type='LoggerHook', interval=50),
param_scheduler=dict(type='ParamSchedulerHook'), param_scheduler=dict(type='ParamSchedulerHook'),

10
configs/beit/upernet_beit-large_fp16_8x1_640x640_160k_ade20k.py
View File

@ -31,10 +31,12 @@ optimizer = dict(
weight_decay=0.05) weight_decay=0.05)
optim_wrapper = dict( optim_wrapper = dict(
type='OptimWrapper', _delete_=True,
type='AmpOptimWrapper',
optimizer=optimizer, optimizer=optimizer,
constructor='LayerDecayOptimizerConstructor', constructor='LayerDecayOptimizerConstructor',
paramwise_cfg=dict(num_layers=24, layer_decay_rate=0.95)) paramwise_cfg=dict(num_layers=24, layer_decay_rate=0.95),
accumulative_counts=2)
param_scheduler = [ param_scheduler = [
dict( dict(
@ -52,7 +54,3 @@ param_scheduler = [
train_dataloader = dict(batch_size=1) train_dataloader = dict(batch_size=1)
val_dataloader = dict(batch_size=1) val_dataloader = dict(batch_size=1)
test_dataloader = val_dataloader test_dataloader = val_dataloader
optimizer_config = dict(
type='GradientCumulativeFp16OptimizerHook', cumulative_iters=2)
fp16 = dict()

10
configs/bisenetv2/bisenetv2_fcn_fp16_4x4_1024x1024_160k_cityscapes.py
View File

@ -1,5 +1,7 @@
_base_ = './bisenetv2_fcn_4x4_1024x1024_160k_cityscapes.py' _base_ = './bisenetv2_fcn_4x4_1024x1024_160k_cityscapes.py'
# fp16 settings optimizer = dict(type='SGD', lr=0.05, momentum=0.9, weight_decay=0.0005)
default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.)) optim_wrapper = dict(
# fp16 placeholder _delete_=True,
fp16 = dict() type='AmpOptimWrapper',
optimizer=optimizer,
loss_scale=512.)

11
configs/convnext/upernet_convnext_base_fp16_512x512_160k_ade20k.py
View File

@ -19,14 +19,16 @@ optimizer = dict(
weight_decay=0.05) weight_decay=0.05)
optim_wrapper = dict( optim_wrapper = dict(
type='OptimWrapper', _delete_=True,
type='AmpOptimWrapper',
optimizer=optimizer, optimizer=optimizer,
paramwise_cfg={ paramwise_cfg={
'decay_rate': 0.9, 'decay_rate': 0.9,
'decay_type': 'stage_wise', 'decay_type': 'stage_wise',
'num_layers': 12 'num_layers': 12
}, },
constructor='LearningRateDecayOptimizerConstructor') constructor='LearningRateDecayOptimizerConstructor',
loss_scale='dynamic')
param_scheduler = [ param_scheduler = [
dict( dict(
@ -45,8 +47,3 @@ param_scheduler = [
train_dataloader = dict(batch_size=2) train_dataloader = dict(batch_size=2)
val_dataloader = dict(batch_size=1) val_dataloader = dict(batch_size=1)
test_dataloader = val_dataloader test_dataloader = val_dataloader
# fp16 settings
default_hooks = dict(
optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
# fp16 placeholder
fp16 = dict()

11
configs/convnext/upernet_convnext_base_fp16_640x640_160k_ade20k.py
View File

@ -34,14 +34,16 @@ optimizer = dict(
weight_decay=0.05) weight_decay=0.05)
optim_wrapper = dict( optim_wrapper = dict(
type='OptimWrapper', _delete_=True,
type='AmpOptimWrapper',
optimizer=optimizer, optimizer=optimizer,
paramwise_cfg={ paramwise_cfg={
'decay_rate': 0.9, 'decay_rate': 0.9,
'decay_type': 'stage_wise', 'decay_type': 'stage_wise',
'num_layers': 12 'num_layers': 12
}, },
constructor='LearningRateDecayOptimizerConstructor') constructor='LearningRateDecayOptimizerConstructor',
loss_scale='dynamic')
param_scheduler = [ param_scheduler = [
dict( dict(
@ -60,8 +62,3 @@ param_scheduler = [
train_dataloader = dict(batch_size=2) train_dataloader = dict(batch_size=2)
val_dataloader = dict(batch_size=1) val_dataloader = dict(batch_size=1)
test_dataloader = val_dataloader test_dataloader = val_dataloader
# fp16 settings
default_hooks = dict(
optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
# fp16 placeholder
fp16 = dict()

11
configs/convnext/upernet_convnext_large_fp16_640x640_160k_ade20k.py
View File

@ -34,14 +34,16 @@ optimizer = dict(
weight_decay=0.05) weight_decay=0.05)
optim_wrapper = dict( optim_wrapper = dict(
type='OptimWrapper', _delete_=True,
type='AmpOptimWrapper',
optimizer=optimizer, optimizer=optimizer,
paramwise_cfg={ paramwise_cfg={
'decay_rate': 0.9, 'decay_rate': 0.9,
'decay_type': 'stage_wise', 'decay_type': 'stage_wise',
'num_layers': 12 'num_layers': 12
}, },
constructor='LearningRateDecayOptimizerConstructor') constructor='LearningRateDecayOptimizerConstructor',
loss_scale='dynamic')
param_scheduler = [ param_scheduler = [
dict( dict(
@ -60,8 +62,3 @@ param_scheduler = [
train_dataloader = dict(batch_size=2) train_dataloader = dict(batch_size=2)
val_dataloader = dict(batch_size=1) val_dataloader = dict(batch_size=1)
test_dataloader = val_dataloader test_dataloader = val_dataloader
# fp16 settings
default_hooks = dict(
optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
# fp16 placeholder
fp16 = dict()

11
configs/convnext/upernet_convnext_small_fp16_512x512_160k_ade20k.py
View File

@ -33,14 +33,16 @@ optimizer = dict(
weight_decay=0.05) weight_decay=0.05)
optim_wrapper = dict( optim_wrapper = dict(
type='OptimWrapper', _delete_=True,
type='AmpOptimWrapper',
optimizer=optimizer, optimizer=optimizer,
paramwise_cfg={ paramwise_cfg={
'decay_rate': 0.9, 'decay_rate': 0.9,
'decay_type': 'stage_wise', 'decay_type': 'stage_wise',
'num_layers': 12 'num_layers': 12
}, },
constructor='LearningRateDecayOptimizerConstructor') constructor='LearningRateDecayOptimizerConstructor',
loss_scale='dynamic')
param_scheduler = [ param_scheduler = [
dict( dict(
@ -59,8 +61,3 @@ param_scheduler = [
train_dataloader = dict(batch_size=2) train_dataloader = dict(batch_size=2)
val_dataloader = dict(batch_size=1) val_dataloader = dict(batch_size=1)
test_dataloader = val_dataloader test_dataloader = val_dataloader
# fp16 settings
default_hooks = dict(
optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
# fp16 placeholder
fp16 = dict()

11
configs/convnext/upernet_convnext_tiny_fp16_512x512_160k_ade20k.py
View File

@ -33,14 +33,16 @@ optimizer = dict(
weight_decay=0.05) weight_decay=0.05)
optim_wrapper = dict( optim_wrapper = dict(
type='OptimWrapper', _delete_=True,
type='AmpOptimWrapper',
optimizer=optimizer, optimizer=optimizer,
paramwise_cfg={ paramwise_cfg={
'decay_rate': 0.9, 'decay_rate': 0.9,
'decay_type': 'stage_wise', 'decay_type': 'stage_wise',
'num_layers': 6 'num_layers': 6
}, },
constructor='LearningRateDecayOptimizerConstructor') constructor='LearningRateDecayOptimizerConstructor',
loss_scale='dynamic')
param_scheduler = [ param_scheduler = [
dict( dict(
@ -59,8 +61,3 @@ param_scheduler = [
train_dataloader = dict(batch_size=2) train_dataloader = dict(batch_size=2)
val_dataloader = dict(batch_size=1) val_dataloader = dict(batch_size=1)
test_dataloader = val_dataloader test_dataloader = val_dataloader
# fp16 settings
default_hooks = dict(
optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
# fp16 placeholder
fp16 = dict()

11
configs/convnext/upernet_convnext_xlarge_fp16_640x640_160k_ade20k.py
View File

@ -34,14 +34,16 @@ optimizer = dict(
weight_decay=0.05) weight_decay=0.05)
optim_wrapper = dict( optim_wrapper = dict(
type='OptimWrapper', _delete_=True,
type='AmpOptimWrapper',
optimizer=optimizer, optimizer=optimizer,
paramwise_cfg={ paramwise_cfg={
'decay_rate': 0.9, 'decay_rate': 0.9,
'decay_type': 'stage_wise', 'decay_type': 'stage_wise',
'num_layers': 12 'num_layers': 12
}, },
constructor='LearningRateDecayOptimizerConstructor') constructor='LearningRateDecayOptimizerConstructor',
loss_scale='dynamic')
param_scheduler = [ param_scheduler = [
dict( dict(
@ -60,8 +62,3 @@ param_scheduler = [
train_dataloader = dict(batch_size=2) train_dataloader = dict(batch_size=2)
val_dataloader = dict(batch_size=1) val_dataloader = dict(batch_size=1)
test_dataloader = val_dataloader test_dataloader = val_dataloader
# fp16 settings
default_hooks = dict(
optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
# fp16 placeholder
fp16 = dict()

10
configs/deeplabv3/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes.py
View File

@ -1,5 +1,7 @@
_base_ = './deeplabv3_r101-d8_512x1024_80k_cityscapes.py' _base_ = './deeplabv3_r101-d8_512x1024_80k_cityscapes.py'
# fp16 settings optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.)) optim_wrapper = dict(
# fp16 placeholder _delete_=True,
fp16 = dict() type='AmpOptimWrapper',
optimizer=optimizer,
loss_scale=512.)

10
configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py
View File

@ -1,5 +1,7 @@
_base_ = './deeplabv3plus_r101-d8_512x1024_80k_cityscapes.py' _base_ = './deeplabv3plus_r101-d8_512x1024_80k_cityscapes.py'
# fp16 settings optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.)) optim_wrapper = dict(
# fp16 placeholder _delete_=True,
fp16 = dict() type='AmpOptimWrapper',
optimizer=optimizer,
loss_scale=512.)

10
configs/fcn/fcn_r101-d8_fp16_512x1024_80k_cityscapes.py
View File

@ -1,5 +1,7 @@
_base_ = './fcn_r101-d8_512x1024_80k_cityscapes.py' _base_ = './fcn_r101-d8_512x1024_80k_cityscapes.py'
# fp16 settings optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.)) optim_wrapper = dict(
# fp16 placeholder _delete_=True,
fp16 = dict() type='AmpOptimWrapper',
optimizer=optimizer,
loss_scale=512.)

11
configs/knet/knet_s3_deeplabv3_r50-d8_8x2_512x512_adamw_80k_ade20k.py
View File

@ -81,12 +81,11 @@ model = dict(
# optimizer # optimizer
optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005) optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(
default_hooks = dict( _delete_=True,
optimizer=dict( type='OptimWrapper',
_delete_=True, optimizer=optimizer,
type='OptimizerHook', clip_grad=dict(max_norm=1, norm_type=2))
grad_clip=dict(max_norm=1, norm_type=2)))
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(

12
configs/knet/knet_s3_fcn_r50-d8_8x2_512x512_adamw_80k_ade20k.py
View File

@ -81,12 +81,12 @@ model = dict(
test_cfg=dict(mode='whole')) test_cfg=dict(mode='whole'))
# optimizer # optimizer
optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005) optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(
default_hooks = dict( _delete_=True,
optimizer=dict( type='OptimWrapper',
_delete_=True, optimizer=optimizer,
type='OptimizerHook', clip_grad=dict(max_norm=1, norm_type=2))
grad_clip=dict(max_norm=1, norm_type=2)))
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(

11
configs/knet/knet_s3_pspnet_r50-d8_8x2_512x512_adamw_80k_ade20k.py
View File

@ -80,12 +80,11 @@ model = dict(
test_cfg=dict(mode='whole')) test_cfg=dict(mode='whole'))
# optimizer # optimizer
optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005) optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(
default_hooks = dict( _delete_=True,
optimizer=dict( type='OptimWrapper',
_delete_=True, optimizer=optimizer,
type='OptimizerHook', clip_grad=dict(max_norm=1, norm_type=2))
grad_clip=dict(max_norm=1, norm_type=2)))
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(

11
configs/knet/knet_s3_upernet_r50-d8_8x2_512x512_adamw_80k_ade20k.py
View File

@ -81,12 +81,11 @@ model = dict(
test_cfg=dict(mode='whole')) test_cfg=dict(mode='whole'))
# optimizer # optimizer
optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005) optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(
default_hooks = dict( _delete_=True,
optimizer=dict( type='OptimWrapper',
_delete_=True, optimizer=optimizer,
type='OptimizerHook', clip_grad=dict(max_norm=1, norm_type=2))
grad_clip=dict(max_norm=1, norm_type=2)))
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(

9
configs/knet/knet_s3_upernet_swin-t_8x2_512x512_adamw_80k_ade20k.py
View File

@ -39,6 +39,7 @@ optimizer = dict(
weight_decay=0.0005) weight_decay=0.0005)
optim_wrapper = dict( optim_wrapper = dict(
_delete_=True,
type='OptimWrapper', type='OptimWrapper',
optimizer=optimizer, optimizer=optimizer,
paramwise_cfg=dict( paramwise_cfg=dict(
@ -46,13 +47,9 @@ optim_wrapper = dict(
'absolute_pos_embed': dict(decay_mult=0.), 'absolute_pos_embed': dict(decay_mult=0.),
'relative_position_bias_table': dict(decay_mult=0.), 'relative_position_bias_table': dict(decay_mult=0.),
'norm': dict(decay_mult=0.) 'norm': dict(decay_mult=0.)
})) }),
clip_grad=dict(max_norm=1, norm_type=2))
default_hooks = dict(
optimizer=dict(
_delete_=True,
type='OptimizerHook',
grad_clip=dict(max_norm=1, norm_type=2)))
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(

10
configs/pspnet/pspnet_r101-d8_fp16_512x1024_80k_cityscapes.py
View File

@ -1,5 +1,7 @@
_base_ = './pspnet_r101-d8_512x1024_80k_cityscapes.py' _base_ = './pspnet_r101-d8_512x1024_80k_cityscapes.py'
# fp16 settings optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.)) optim_wrapper = dict(
# fp16 placeholder _delete_=True,
fp16 = dict() type='AmpOptimWrapper',
optimizer=optimizer,
loss_scale=512.)

11
configs/pspnet/pspnet_r50-d32_rsb-pretrain_512x1024_adamw_80k_cityscapes.py
View File

@ -16,12 +16,11 @@ model = dict(
strides=(1, 2, 2, 2))) strides=(1, 2, 2, 2)))
optimizer = dict(_delete_=True, type='AdamW', lr=0.0005, weight_decay=0.05) optimizer = dict(_delete_=True, type='AdamW', lr=0.0005, weight_decay=0.05)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(
default_hooks = dict( _delete_=True,
optimizer=dict( type='OptimWrapper',
_delete_=True, optimizer=optimizer,
type='OptimizerHook', clip_grad=dict(max_norm=1, norm_type=2))
grad_clip=dict(max_norm=1, norm_type=2)))
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(

11
configs/pspnet/pspnet_r50-d8_rsb-pretrain_512x1024_adamw_80k_cityscapes.py
View File

@ -14,12 +14,11 @@ model = dict(
type='Pretrained', prefix='backbone.', checkpoint=checkpoint))) type='Pretrained', prefix='backbone.', checkpoint=checkpoint)))
optimizer = dict(_delete_=True, type='AdamW', lr=0.0005, weight_decay=0.05) optimizer = dict(_delete_=True, type='AdamW', lr=0.0005, weight_decay=0.05)
optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer) optim_wrapper = dict(
default_hooks = dict( _delete_=True,
optimizer=dict( type='OptimWrapper',
_delete_=True, optimizer=optimizer,
type='OptimizerHook', clip_grad=dict(max_norm=1, norm_type=2))
grad_clip=dict(max_norm=1, norm_type=2)))
# learning policy # learning policy
param_scheduler = [ param_scheduler = [
dict( dict(