[Refactor] Refacor default_hooks and train & val & test loops in configs

configs/_base_/default_runtime.py

@@ -1,14 +1,9 @@
-# yapf:disable
+default_scope = 'mmseg'
-log_config = dict(
+env_cfg = dict(
-    interval=50,
+    cudnn_benchmark=True,
-    hooks=[
+    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
-        dict(type='TextLoggerHook', by_epoch=False),
+    dist_cfg=dict(backend='nccl'),
-        # dict(type='TensorboardLoggerHook')
+)
-    ])
-# yapf:enable
-dist_params = dict(backend='nccl')
 log_level = 'INFO'
 load_from = None
-resume_from = None
+resume = False
-workflow = [('train', 1)]
-cudnn_benchmark = True

configs/_base_/schedules/schedule_160k.py

@@ -1,9 +1,19 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
-optimizer_config = dict()
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 # learning policy
 lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
-# runtime settings
+# training schedule for 160k
-runner = dict(type='IterBasedRunner', max_iters=160000)
+train_cfg = dict(
-checkpoint_config = dict(by_epoch=False, interval=16000)
+    type='IterBasedTrainLoop', max_iters=160000, val_interval=16000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
 evaluation = dict(interval=16000, metric='mIoU', pre_eval=True)
+default_hooks = dict(
+    optimizer=dict(type='OptimizerHook', grad_clip=None),
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=50),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=16000),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+)

configs/_base_/schedules/schedule_20k.py

@@ -1,9 +1,18 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
-optimizer_config = dict()
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 # learning policy
 lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
-# runtime settings
+# training schedule for 20k
-runner = dict(type='IterBasedRunner', max_iters=20000)
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=20000, val_interval=2000)
-checkpoint_config = dict(by_epoch=False, interval=2000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
 evaluation = dict(interval=2000, metric='mIoU', pre_eval=True)
+default_hooks = dict(
+    optimizer=dict(type='OptimizerHook', grad_clip=None),
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=50),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=2000),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+)

configs/_base_/schedules/schedule_320k.py

@@ -1,9 +1,19 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
-optimizer_config = dict()
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 # learning policy
 lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
-# runtime settings
+# training schedule for 320k
-runner = dict(type='IterBasedRunner', max_iters=320000)
+train_cfg = dict(
-checkpoint_config = dict(by_epoch=False, interval=32000)
+    type='IterBasedTrainLoop', max_iters=320000, val_interval=32000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
 evaluation = dict(interval=32000, metric='mIoU')
+default_hooks = dict(
+    optimizer=dict(type='OptimizerHook', grad_clip=None),
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=50),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=32000),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+)

configs/_base_/schedules/schedule_40k.py

@@ -1,9 +1,18 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
-optimizer_config = dict()
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 # learning policy
 lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
-# runtime settings
+# training schedule for 40k
-runner = dict(type='IterBasedRunner', max_iters=40000)
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=40000, val_interval=4000)
-checkpoint_config = dict(by_epoch=False, interval=4000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
 evaluation = dict(interval=4000, metric='mIoU', pre_eval=True)
+default_hooks = dict(
+    optimizer=dict(type='OptimizerHook', grad_clip=None),
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=50),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=4000),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+)

configs/_base_/schedules/schedule_80k.py

@@ -1,9 +1,18 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
-optimizer_config = dict()
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 # learning policy
 lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
-# runtime settings
+# training schedule for 80k
-runner = dict(type='IterBasedRunner', max_iters=80000)
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=80000, val_interval=8000)
-checkpoint_config = dict(by_epoch=False, interval=8000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
 evaluation = dict(interval=8000, metric='mIoU', pre_eval=True)
+default_hooks = dict(
+    optimizer=dict(type='OptimizerHook', grad_clip=None),
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=50),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=8000),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+)

configs/beit/upernet_beit-base_8x2_640x640_160k_ade20k.py

@@ -12,7 +12,11 @@ optimizer = dict(
     type='AdamW',
     lr=3e-5,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     constructor='LayerDecayOptimizerConstructor',
     paramwise_cfg=dict(num_layers=12, layer_decay_rate=0.9))
 

configs/beit/upernet_beit-large_fp16_8x1_640x640_160k_ade20k.py

@@ -26,10 +26,13 @@ optimizer = dict(
     type='AdamW',
     lr=2e-5,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     constructor='LayerDecayOptimizerConstructor',
     paramwise_cfg=dict(num_layers=24, layer_decay_rate=0.95))
-
 lr_config = dict(
     _delete_=True,
     policy='poly',

configs/bisenetv1/bisenetv1_r101-d32_lr5e-3_4x4_512x512_160k_coco-stuff164k.py

@@ -15,4 +15,8 @@ model = dict(
         dict(in_channels=512, channels=256, num_classes=171),
     ])
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.005)
+optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
+train_dataloader = dict(batch_size=4, num_workers=4)
+val_dataloader = dict(batch_size=4, num_workers=4)
+test_dataloader = val_dataloader

configs/bisenetv1/bisenetv1_r18-d32_4x4_1024x1024_160k_cityscapes.py

@@ -4,7 +4,8 @@ _base_ = [
     '../_base_/default_runtime.py', '../_base_/schedules/schedule_160k.py'
 ]
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.025)
+optimizer = dict(type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 train_dataloader = dict(batch_size=4, num_workers=4)
 val_dataloader = dict(batch_size=4, num_workers=4)
 test_dataloader = val_dataloader

configs/bisenetv1/bisenetv1_r18-d32_in1k-pre_4x4_1024x1024_160k_cityscapes.py

@@ -9,7 +9,8 @@ model = dict(
             init_cfg=dict(
                 type='Pretrained', checkpoint='open-mmlab://resnet18_v1c'))))
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.025)
+optimizer = dict(type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 train_dataloader = dict(batch_size=4, num_workers=4)
 val_dataloader = dict(batch_size=4, num_workers=4)
 test_dataloader = val_dataloader

configs/bisenetv1/bisenetv1_r18-d32_lr5e-3_4x4_512x512_160k_coco-stuff164k.py

@@ -10,4 +10,5 @@ model = dict(
         dict(num_classes=171),
     ])
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.005)
+optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)

configs/bisenetv1/bisenetv1_r50-d32_4x4_1024x1024_160k_cityscapes.py

@@ -35,7 +35,8 @@ model = dict(
             concat_input=False),
     ])
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.05)
+optimizer = dict(type='SGD', lr=0.05, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 train_dataloader = dict(batch_size=4, num_workers=4)
 val_dataloader = dict(batch_size=4, num_workers=4)
 test_dataloader = val_dataloader

configs/bisenetv1/bisenetv1_r50-d32_lr5e-3_4x4_512x512_160k_coco-stuff164k.py

@@ -15,4 +15,5 @@ model = dict(
         dict(in_channels=512, channels=256, num_classes=171),
     ])
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.005)
+optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)

configs/bisenetv2/bisenetv2_fcn_4x4_1024x1024_160k_cityscapes.py

@@ -4,7 +4,8 @@ _base_ = [
     '../_base_/default_runtime.py', '../_base_/schedules/schedule_160k.py'
 ]
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.05)
+optimizer = dict(type='SGD', lr=0.05, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 train_dataloader = dict(batch_size=4, num_workers=4)
 val_dataloader = dict(batch_size=4, num_workers=4)
 test_dataloader = val_dataloader

configs/bisenetv2/bisenetv2_fcn_4x8_1024x1024_160k_cityscapes.py

@@ -4,7 +4,8 @@ _base_ = [
     '../_base_/default_runtime.py', '../_base_/schedules/schedule_160k.py'
 ]
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.05)
+optimizer = dict(type='SGD', lr=0.05, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 train_dataloader = dict(batch_size=8, num_workers=4)
 val_dataloader = dict(batch_size=8, num_workers=4)
 test_dataloader = val_dataloader

configs/bisenetv2/bisenetv2_fcn_fp16_4x4_1024x1024_160k_cityscapes.py

@@ -1,5 +1,5 @@
 _base_ = './bisenetv2_fcn_4x4_1024x1024_160k_cityscapes.py'
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale=512.)
+default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.))
 # fp16 placeholder
 fp16 = dict()

configs/bisenetv2/bisenetv2_fcn_ohem_4x4_1024x1024_160k_cityscapes.py

@@ -3,9 +3,68 @@ _base_ = [
     '../_base_/datasets/cityscapes_1024x1024.py',
     '../_base_/default_runtime.py', '../_base_/schedules/schedule_160k.py'
 ]
-sampler = dict(type='OHEMPixelSampler', thresh=0.7, min_kept=10000)
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+models = dict(
+    decode_head=dict(
+        sampler=dict(type='OHEMPixelSampler', thresh=0.7, min_kept=10000)),
+    auxiliary_head=[
+        dict(
+            type='FCNHead',
+            in_channels=16,
+            channels=16,
+            num_convs=2,
+            num_classes=19,
+            in_index=1,
+            norm_cfg=norm_cfg,
+            concat_input=False,
+            align_corners=False,
+            sampler=dict(type='OHEMPixelSampler', thresh=0.7, min_kept=10000),
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+        dict(
+            type='FCNHead',
+            in_channels=32,
+            channels=64,
+            num_convs=2,
+            num_classes=19,
+            in_index=2,
+            norm_cfg=norm_cfg,
+            concat_input=False,
+            align_corners=False,
+            sampler=dict(type='OHEMPixelSampler', thresh=0.7, min_kept=10000),
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+        dict(
+            type='FCNHead',
+            in_channels=64,
+            channels=256,
+            num_convs=2,
+            num_classes=19,
+            in_index=3,
+            norm_cfg=norm_cfg,
+            concat_input=False,
+            align_corners=False,
+            sampler=dict(type='OHEMPixelSampler', thresh=0.7, min_kept=10000),
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+        dict(
+            type='FCNHead',
+            in_channels=128,
+            channels=1024,
+            num_convs=2,
+            num_classes=19,
+            in_index=4,
+            norm_cfg=norm_cfg,
+            concat_input=False,
+            align_corners=False,
+            sampler=dict(type='OHEMPixelSampler', thresh=0.7, min_kept=10000),
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    ],
+)
 lr_config = dict(warmup='linear', warmup_iters=1000)
-optimizer = dict(lr=0.05)
+optimizer = dict(type='SGD', lr=0.05, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 train_dataloader = dict(batch_size=4, num_workers=4)
 val_dataloader = dict(batch_size=4, num_workers=4)
 test_dataloader = val_dataloader

configs/cgnet/cgnet_512x1024_60k_cityscapes.py

@@ -2,12 +2,16 @@ _base_ = ['../_base_/models/cgnet.py', '../_base_/default_runtime.py']
 
 # optimizer
 optimizer = dict(type='Adam', lr=0.001, eps=1e-08, weight_decay=0.0005)
-optimizer_config = dict()
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 # learning policy
 lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
 # runtime settings
 total_iters = 60000
-checkpoint_config = dict(by_epoch=False, interval=4000)
+train_cfg = dict(
+    type='IterBasedTrainLoop', max_iters=total_iters, val_interval=4000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(checkpoint=dict(by_epoch=False, interval=4000))
 evaluation = dict(interval=4000, metric='mIoU')
 
 # dataset settings

configs/cgnet/cgnet_680x680_60k_cityscapes.py

@@ -5,12 +5,16 @@ _base_ = [
 
 # optimizer
 optimizer = dict(type='Adam', lr=0.001, eps=1e-08, weight_decay=0.0005)
-optimizer_config = dict()
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 # learning policy
 lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
 # runtime settings
 total_iters = 60000
-checkpoint_config = dict(by_epoch=False, interval=4000)
+train_cfg = dict(
+    type='IterBasedTrainLoop', max_iters=total_iters, val_interval=4000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(checkpoint=dict(by_epoch=False, interval=4000))
 evaluation = dict(interval=4000, metric='mIoU')
 
 img_norm_cfg = dict(

configs/convnext/upernet_convnext_base_fp16_512x512_160k_ade20k.py

@@ -10,17 +10,21 @@ model = dict(
 )
 
 optimizer = dict(
-    constructor='LearningRateDecayOptimizerConstructor',
     _delete_=True,
     type='AdamW',
     lr=0.0001,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg={
         'decay_rate': 0.9,
         'decay_type': 'stage_wise',
         'num_layers': 12
-    })
+    },
+    constructor='LearningRateDecayOptimizerConstructor')
 
 lr_config = dict(
     _delete_=True,
@@ -37,6 +41,7 @@ train_dataloader = dict(batch_size=2)
 val_dataloader = dict(batch_size=2)
 test_dataloader = val_dataloader
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale='dynamic')
+default_hooks = dict(
+    optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
 # fp16 placeholder
 fp16 = dict()

configs/convnext/upernet_convnext_base_fp16_640x640_160k_ade20k.py

@@ -25,17 +25,21 @@ model = dict(
 )
 
 optimizer = dict(
-    constructor='LearningRateDecayOptimizerConstructor',
     _delete_=True,
     type='AdamW',
     lr=0.0001,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg={
         'decay_rate': 0.9,
         'decay_type': 'stage_wise',
         'num_layers': 12
-    })
+    },
+    constructor='LearningRateDecayOptimizerConstructor')
 
 lr_config = dict(
     _delete_=True,
@@ -52,6 +56,7 @@ train_dataloader = dict(batch_size=2)
 val_dataloader = dict(batch_size=2)
 test_dataloader = val_dataloader
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale='dynamic')
+default_hooks = dict(
+    optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
 # fp16 placeholder
 fp16 = dict()

configs/convnext/upernet_convnext_large_fp16_640x640_160k_ade20k.py

@@ -25,17 +25,21 @@ model = dict(
 )
 
 optimizer = dict(
-    constructor='LearningRateDecayOptimizerConstructor',
     _delete_=True,
     type='AdamW',
     lr=0.0001,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg={
         'decay_rate': 0.9,
         'decay_type': 'stage_wise',
         'num_layers': 12
-    })
+    },
+    constructor='LearningRateDecayOptimizerConstructor')
 
 lr_config = dict(
     _delete_=True,
@@ -52,6 +56,7 @@ train_dataloader = dict(batch_size=2)
 val_dataloader = dict(batch_size=2)
 test_dataloader = val_dataloader
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale='dynamic')
+default_hooks = dict(
+    optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
 # fp16 placeholder
 fp16 = dict()

configs/convnext/upernet_convnext_small_fp16_512x512_160k_ade20k.py

@@ -24,17 +24,21 @@ model = dict(
 )
 
 optimizer = dict(
-    constructor='LearningRateDecayOptimizerConstructor',
     _delete_=True,
     type='AdamW',
     lr=0.0001,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg={
         'decay_rate': 0.9,
         'decay_type': 'stage_wise',
         'num_layers': 12
-    })
+    },
+    constructor='LearningRateDecayOptimizerConstructor')
 
 lr_config = dict(
     _delete_=True,
@@ -51,6 +55,7 @@ train_dataloader = dict(batch_size=2)
 val_dataloader = dict(batch_size=2)
 test_dataloader = val_dataloader
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale='dynamic')
+default_hooks = dict(
+    optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
 # fp16 placeholder
 fp16 = dict()

configs/convnext/upernet_convnext_tiny_fp16_512x512_160k_ade20k.py

@@ -24,17 +24,21 @@ model = dict(
 )
 
 optimizer = dict(
-    constructor='LearningRateDecayOptimizerConstructor',
     _delete_=True,
     type='AdamW',
     lr=0.0001,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg={
         'decay_rate': 0.9,
         'decay_type': 'stage_wise',
         'num_layers': 6
-    })
+    },
+    constructor='LearningRateDecayOptimizerConstructor')
 
 lr_config = dict(
     _delete_=True,
@@ -51,6 +55,7 @@ train_dataloader = dict(batch_size=2)
 val_dataloader = dict(batch_size=2)
 test_dataloader = val_dataloader
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale='dynamic')
+default_hooks = dict(
+    optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
 # fp16 placeholder
 fp16 = dict()

configs/convnext/upernet_convnext_xlarge_fp16_640x640_160k_ade20k.py

@@ -25,17 +25,21 @@ model = dict(
 )
 
 optimizer = dict(
-    constructor='LearningRateDecayOptimizerConstructor',
     _delete_=True,
     type='AdamW',
     lr=0.00008,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg={
         'decay_rate': 0.9,
         'decay_type': 'stage_wise',
         'num_layers': 12
-    })
+    },
+    constructor='LearningRateDecayOptimizerConstructor')
 
 lr_config = dict(
     _delete_=True,
@@ -52,6 +56,7 @@ train_dataloader = dict(batch_size=2)
 val_dataloader = dict(batch_size=2)
 test_dataloader = val_dataloader
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale='dynamic')
+default_hooks = dict(
+    optimizer=dict(type='Fp16OptimizerHook', loss_scale='dynamic'))
 # fp16 placeholder
 fp16 = dict()

configs/deeplabv3/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes.py

@@ -1,5 +1,5 @@
 _base_ = './deeplabv3_r101-d8_512x1024_80k_cityscapes.py'
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale=512.)
+default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.))
 # fp16 placeholder
 fp16 = dict()

configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py

@@ -1,5 +1,5 @@
 _base_ = './deeplabv3plus_r101-d8_512x1024_80k_cityscapes.py'
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale=512.)
+default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.))
 # fp16 placeholder
 fp16 = dict()

configs/dpt/dpt_vit-b16_512x512_160k_ade20k.py

@@ -10,7 +10,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'pos_embed': dict(decay_mult=0.),

configs/fcn/fcn_r101-d8_fp16_512x1024_80k_cityscapes.py

@@ -1,5 +1,5 @@
 _base_ = './fcn_r101-d8_512x1024_80k_cityscapes.py'
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale=512.)
+default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.))
 # fp16 placeholder
 fp16 = dict()

configs/knet/knet_s3_deeplabv3_r50-d8_8x2_512x512_adamw_80k_ade20k.py

@@ -79,7 +79,12 @@ model = dict(
 
 # optimizer
 optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
-optimizer_config = dict(grad_clip=dict(max_norm=1, norm_type=2))
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
+default_hooks = dict(
+    optimizer=dict(
+        _delete_=True,
+        type='OptimizerHook',
+        grad_clip=dict(max_norm=1, norm_type=2)))
 # learning policy
 lr_config = dict(
     _delete_=True,

configs/knet/knet_s3_fcn_r50-d8_8x2_512x512_adamw_80k_ade20k.py

@@ -79,7 +79,12 @@ model = dict(
     test_cfg=dict(mode='whole'))
 # optimizer
 optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
-optimizer_config = dict(grad_clip=dict(max_norm=1, norm_type=2))
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
+default_hooks = dict(
+    optimizer=dict(
+        _delete_=True,
+        type='OptimizerHook',
+        grad_clip=dict(max_norm=1, norm_type=2)))
 # learning policy
 lr_config = dict(
     _delete_=True,

configs/knet/knet_s3_pspnet_r50-d8_8x2_512x512_adamw_80k_ade20k.py

@@ -78,7 +78,12 @@ model = dict(
     test_cfg=dict(mode='whole'))
 # optimizer
 optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
-optimizer_config = dict(grad_clip=dict(max_norm=1, norm_type=2))
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
+default_hooks = dict(
+    optimizer=dict(
+        _delete_=True,
+        type='OptimizerHook',
+        grad_clip=dict(max_norm=1, norm_type=2)))
 # learning policy
 lr_config = dict(
     _delete_=True,

configs/knet/knet_s3_upernet_r50-d8_8x2_512x512_adamw_80k_ade20k.py

@@ -79,7 +79,12 @@ model = dict(
     test_cfg=dict(mode='whole'))
 # optimizer
 optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0005)
-optimizer_config = dict(grad_clip=dict(max_norm=1, norm_type=2))
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
+default_hooks = dict(
+    optimizer=dict(
+        _delete_=True,
+        type='OptimizerHook',
+        grad_clip=dict(max_norm=1, norm_type=2)))
 # learning policy
 lr_config = dict(
     _delete_=True,

configs/knet/knet_s3_upernet_swin-t_8x2_512x512_adamw_80k_ade20k.py

@@ -36,14 +36,23 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.0005,
+    weight_decay=0.0005)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'absolute_pos_embed': dict(decay_mult=0.),
             'relative_position_bias_table': dict(decay_mult=0.),
             'norm': dict(decay_mult=0.)
         }))
-optimizer_config = dict(grad_clip=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
 lr_config = dict(
     _delete_=True,

configs/mae/upernet_mae-base_fp16_8x2_512x512_160k_ade20k.py

@@ -27,9 +27,13 @@ optimizer = dict(
     type='AdamW',
     lr=1e-4,
     betas=(0.9, 0.999),
-    weight_decay=0.05,
+    weight_decay=0.05)
-    constructor='LayerDecayOptimizerConstructor',
+
-    paramwise_cfg=dict(num_layers=12, layer_decay_rate=0.65))
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
+    paramwise_cfg=dict(num_layers=12, layer_decay_rate=0.65),
+    constructor='LayerDecayOptimizerConstructor')
 
 lr_config = dict(
     _delete_=True,

configs/ocrnet/ocrnet_r101-d8_512x1024_40k_b16_cityscapes.py

@@ -3,5 +3,6 @@ _base_ = [
     '../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
 ]
 model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))
-optimizer = dict(lr=0.02)
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 lr_config = dict(min_lr=2e-4)

configs/ocrnet/ocrnet_r101-d8_512x1024_80k_b16_cityscapes.py

@@ -3,5 +3,6 @@ _base_ = [
     '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
 ]
 model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))
-optimizer = dict(lr=0.02)
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
 lr_config = dict(min_lr=2e-4)

configs/pspnet/pspnet_r101-d8_fp16_512x1024_80k_cityscapes.py

@@ -1,5 +1,5 @@
 _base_ = './pspnet_r101-d8_512x1024_80k_cityscapes.py'
 # fp16 settings
-optimizer_config = dict(type='Fp16OptimizerHook', loss_scale=512.)
+default_hooks = dict(optimizer=dict(type='Fp16OptimizerHook', loss_scale=512.))
 # fp16 placeholder
 fp16 = dict()

configs/pspnet/pspnet_r50-d32_rsb-pretrain_512x1024_adamw_80k_cityscapes.py

@@ -13,7 +13,12 @@ model = dict(
         strides=(1, 2, 2, 2)))
 
 optimizer = dict(_delete_=True, type='AdamW', lr=0.0005, weight_decay=0.05)
-optimizer_config = dict(grad_clip=dict(max_norm=1, norm_type=2))
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
+default_hooks = dict(
+    optimizer=dict(
+        _delete_=True,
+        type='OptimizerHook',
+        grad_clip=dict(max_norm=1, norm_type=2)))
 # learning policy
 lr_config = dict(
     _delete_=True,

configs/pspnet/pspnet_r50-d8_rsb-pretrain_512x1024_adamw_80k_cityscapes.py

@@ -11,7 +11,12 @@ model = dict(
             type='Pretrained', prefix='backbone.', checkpoint=checkpoint)))
 
 optimizer = dict(_delete_=True, type='AdamW', lr=0.0005, weight_decay=0.05)
-optimizer_config = dict(grad_clip=dict(max_norm=1, norm_type=2))
+optim_wrapper = dict(type='OptimWrapper', optimizer=optimizer)
+default_hooks = dict(
+    optimizer=dict(
+        _delete_=True,
+        type='OptimizerHook',
+        grad_clip=dict(max_norm=1, norm_type=2)))
 # learning policy
 lr_config = dict(
     _delete_=True,

configs/segformer/segformer_mit-b0_512x512_160k_ade20k.py

@@ -12,7 +12,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'pos_block': dict(decay_mult=0.),

configs/segformer/segformer_mit-b0_8x1_1024x1024_160k_cityscapes.py

@@ -15,14 +15,17 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'pos_block': dict(decay_mult=0.),
             'norm': dict(decay_mult=0.),
             'head': dict(lr_mult=10.)
         }))
-
 lr_config = dict(
     _delete_=True,
     policy='poly',

configs/swin/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K.py

@@ -23,7 +23,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'absolute_pos_embed': dict(decay_mult=0.),

configs/twins/twins_pcpvt-s_uperhead_8x4_512x512_160k_ade20k.py

@@ -9,7 +9,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(custom_keys={
         'pos_block': dict(decay_mult=0.),
         'norm': dict(decay_mult=0.)

configs/twins/twins_svt-s_uperhead_8x2_512x512_160k_ade20k.py

@@ -24,7 +24,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(custom_keys={
         'pos_block': dict(decay_mult=0.),
         'norm': dict(decay_mult=0.)

configs/vit/upernet_vit-b16_ln_mln_512x512_160k_ade20k.py

@@ -17,7 +17,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'pos_embed': dict(decay_mult=0.),

configs/vit/upernet_vit-b16_mln_512x512_160k_ade20k.py

@@ -16,7 +16,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'pos_embed': dict(decay_mult=0.),

configs/vit/upernet_vit-b16_mln_512x512_80k_ade20k.py

@@ -16,7 +16,11 @@ optimizer = dict(
     type='AdamW',
     lr=0.00006,
     betas=(0.9, 0.999),
-    weight_decay=0.01,
+    weight_decay=0.01)
+
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=optimizer,
     paramwise_cfg=dict(
         custom_keys={
             'pos_embed': dict(decay_mult=0.),