Migrate configs to new styles

configs/_base_/default_runtime.py

@@ -1,17 +1,21 @@
-# yapf:disable
+# custom_imports = dict(imports=['mmcv.transforms'], allow_failed_imports=False) # noqa
-log_config = dict(
+default_scope = 'mmocr'
-    interval=5,
+
-    hooks=[
+default_hooks = dict(
-        dict(type='TextLoggerHook')
+    optimizer=dict(type='OptimizerHook', grad_clip=None),
-    ])
+    timer=dict(type='IterTimerHook'),
-# yapf:enable
+    logger=dict(type='LoggerHook', interval=5),
-dist_params = dict(backend='nccl')
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', interval=1),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+)
+
+env_cfg = dict(
+    cudnn_benchmark=True,
+    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
+    dist_cfg=dict(backend='nccl'),
+)
+
 log_level = 'INFO'
 load_from = None
-resume_from = None
+resume = False
-workflow = [('train', 1)]
-
-# disable opencv multithreading to avoid system being overloaded
-opencv_num_threads = 0
-# set multi-process start method as `fork` to speed up the training
-mp_start_method = 'fork'

configs/_base_/schedules/schedule_adadelta_18e.py

@@ -1,8 +0,0 @@
-# optimizer
-optimizer = dict(type='Adadelta', lr=0.5)
-optimizer_config = dict(grad_clip=dict(max_norm=0.5))
-# learning policy
-lr_config = dict(policy='step', step=[8, 14, 16])
-# running settings
-runner = dict(type='EpochBasedRunner', max_epochs=18)
-checkpoint_config = dict(interval=1)

configs/_base_/schedules/schedule_adadelta_5e.py

@@ -1,8 +1,10 @@
-# optimizer
 optimizer = dict(type='Adadelta', lr=1.0)
-optimizer_config = dict(grad_clip=None)
+
-# learning policy
+train_cfg = dict(by_epoch=True, max_epochs=5)
-lr_config = dict(policy='step', step=[])
+val_cfg = dict(interval=1)
-# running settings
+test_cfg = dict()
-runner = dict(type='EpochBasedRunner', max_epochs=5)
+
-checkpoint_config = dict(interval=1)
+# learning rate
+param_scheduler = [
+    dict(type='ConstantLR'),
+]

configs/_base_/schedules/schedule_adam_600e.py

@@ -1,8 +1,11 @@
 # optimizer
 optimizer = dict(type='Adam', lr=1e-3)
-optimizer_config = dict(grad_clip=None)
+
-# learning policy
+train_cfg = dict(by_epoch=True, max_epochs=600)
-lr_config = dict(policy='poly', power=0.9)
+val_cfg = dict(interval=20)
-# running settings
+test_cfg = dict()
-runner = dict(type='EpochBasedRunner', max_epochs=600)
+
-checkpoint_config = dict(interval=100)
+# learning rate
+param_scheduler = [
+    dict(type='PolyLR', power=0.9, end=600),
+]

configs/_base_/schedules/schedule_adam_step_12e.py

@@ -1,12 +1,12 @@
 # optimizer
 optimizer = dict(type='Adam', lr=4e-4)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=True, max_epochs=12)
+val_cfg = dict(interval=1)
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(
+param_scheduler = [
-    policy='step',
+    dict(type='LinearLR', end=100, by_epoch=False),
-    warmup='linear',
+    dict(type='MultiStepLR', milestones=[11], end=12),
-    warmup_iters=100,
+]
-    warmup_ratio=1.0 / 3,
-    step=[11])
-runner = dict(type='EpochBasedRunner', max_epochs=12)
-checkpoint_config = dict(interval=1)

configs/_base_/schedules/schedule_adam_step_20e.py

@@ -1,14 +1,12 @@
 # optimizer
 optimizer = dict(type='Adam', lr=1e-4)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=True, max_epochs=20)
+val_cfg = dict(interval=1)
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(
+param_scheduler = [
-    policy='step',
+    dict(type='LinearLR', end=1, start_factor=0.001),
-    step=[16, 18],
+    dict(type='MultiStepLR', milestones=[16, 18], end=20),
-    warmup='linear',
+]
-    warmup_iters=1,
-    warmup_ratio=0.001,
-    warmup_by_epoch=True)
-# running settings
-runner = dict(type='EpochBasedRunner', max_epochs=20)
-checkpoint_config = dict(interval=1)

configs/_base_/schedules/schedule_adam_step_5e.py

@@ -1,8 +1,11 @@
 # optimizer
 optimizer = dict(type='Adam', lr=1e-3)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=True, max_epochs=5)
+val_cfg = dict(interval=1)
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(policy='step', step=[3, 4])
+param_scheduler = [
-# running settings
+    dict(type='MultiStepLR', milestones=[3, 4], end=5),
-runner = dict(type='EpochBasedRunner', max_epochs=5)
+]
-checkpoint_config = dict(interval=1)

configs/_base_/schedules/schedule_adam_step_600e.py

@@ -1,8 +1,11 @@
 # optimizer
 optimizer = dict(type='Adam', lr=1e-4)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=True, max_epochs=600)
+val_cfg = dict(interval=40)
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(policy='step', step=[200, 400])
+param_scheduler = [
-# running settings
+    dict(type='MultiStepLR', milestones=[200, 400], end=600),
-runner = dict(type='EpochBasedRunner', max_epochs=600)
+]
-checkpoint_config = dict(interval=100)

configs/_base_/schedules/schedule_adam_step_6e.py

@@ -1,8 +1,8 @@
-# optimizer
+_base_ = 'schedule_adam_step_5e.py'
-optimizer = dict(type='Adam', lr=1e-3)
+
-optimizer_config = dict(grad_clip=None)
+train_cfg = dict(by_epoch=True, max_epochs=6)
+
 # learning policy
-lr_config = dict(policy='step', step=[3, 4])
+param_scheduler = [
-# running settings
+    dict(type='MultiStepLR', milestones=[3, 4], end=6),
-runner = dict(type='EpochBasedRunner', max_epochs=6)
+]
-checkpoint_config = dict(interval=1)

configs/_base_/schedules/schedule_sgd_100k_iters.py

@@ -1,8 +1,11 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.007, momentum=0.9, weight_decay=0.0001)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=False, max_iters=100000)
+val_cfg = dict(interval=100001)  # Never evaluate
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(policy='poly', power=0.9, min_lr=1e-7, by_epoch=False)
+param_scheduler = [
-# running settings
+    dict(type='PolyLR', power=0.9, eta_min=1e-7, by_epoch=False, end=100000),
-runner = dict(type='IterBasedRunner', max_iters=100000)
+]
-checkpoint_config = dict(interval=10000)

configs/_base_/schedules/schedule_sgd_1200e.py

@@ -1,8 +1,11 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.007, momentum=0.9, weight_decay=0.0001)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=True, max_epochs=1200)
+val_cfg = dict(interval=20)  # Never evaluate
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(policy='poly', power=0.9, min_lr=1e-7, by_epoch=True)
+param_scheduler = [
-# running settings
+    dict(type='PolyLR', power=0.9, eta_min=1e-7, end=1200),
-runner = dict(type='EpochBasedRunner', max_epochs=1200)
+]
-checkpoint_config = dict(interval=100)

configs/_base_/schedules/schedule_sgd_1500e.py

@@ -1,8 +1,8 @@
-# optimizer
+train_cfg = dict(by_epoch=True, max_epochs=1500)
-optimizer = dict(type='SGD', lr=1e-3, momentum=0.90, weight_decay=5e-4)
+val_cfg = dict(interval=20)  # Never evaluate
-optimizer_config = dict(grad_clip=None)
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(policy='poly', power=0.9, min_lr=1e-7, by_epoch=True)
+param_scheduler = [
-# running settings
+    dict(type='PolyLR', power=0.9, eta_min=1e-7, end=1500),
-runner = dict(type='EpochBasedRunner', max_epochs=1500)
+]
-checkpoint_config = dict(interval=100)

configs/_base_/schedules/schedule_sgd_160e.py

@@ -1,13 +1,12 @@
 # optimizer
 optimizer = dict(type='SGD', lr=0.08, momentum=0.9, weight_decay=0.0001)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=True, max_epochs=160)
+val_cfg = dict(interval=20)
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(
+param_scheduler = [
-    policy='step',
+    dict(type='LinearLR', end=500, start_factor=0.001, by_epoch=False),
-    warmup='linear',
+    dict(type='MultiStepLR', milestones=[80, 128], end=160),
-    warmup_iters=500,
+]
-    warmup_ratio=0.001,
-    step=[80, 128])
-# running settings
-runner = dict(type='EpochBasedRunner', max_epochs=160)
-checkpoint_config = dict(interval=10)

configs/_base_/schedules/schedule_sgd_600e.py

@@ -1,8 +1,10 @@
-# optimizer
 optimizer = dict(type='SGD', lr=1e-3, momentum=0.99, weight_decay=5e-4)
-optimizer_config = dict(grad_clip=None)
+
+train_cfg = dict(by_epoch=True, max_epochs=600)
+val_cfg = dict(interval=50)
+test_cfg = dict()
+
 # learning policy
-lr_config = dict(policy='step', step=[200, 400])
+param_scheduler = [
-# running settings
+    dict(type='MultiStepLR', milestones=[200, 400], end=600),
-runner = dict(type='EpochBasedRunner', max_epochs=600)
+]
-checkpoint_config = dict(interval=100)