mirror of https://github.com/open-mmlab/mmocr.git
55 lines
1.5 KiB
Python
55 lines
1.5 KiB
Python
_base_ = [
|
|
'_base_fcenet_resnet50-dcnv2_fpn.py',
|
|
'../_base_/datasets/ctw1500.py',
|
|
'../_base_/default_runtime.py',
|
|
'../_base_/schedules/schedule_sgd_base.py',
|
|
]
|
|
|
|
optim_wrapper = dict(optimizer=dict(lr=1e-3, weight_decay=5e-4))
|
|
train_cfg = dict(max_epochs=1500)
|
|
# learning policy
|
|
param_scheduler = [
|
|
dict(type='PolyLR', power=0.9, eta_min=1e-7, end=1500),
|
|
]
|
|
|
|
# dataset settings
|
|
ctw1500_textdet_train = _base_.ctw1500_textdet_train
|
|
ctw1500_textdet_test = _base_.ctw1500_textdet_test
|
|
|
|
# test pipeline for CTW1500
|
|
ctw_test_pipeline = [
|
|
dict(type='LoadImageFromFile', color_type='color_ignore_orientation'),
|
|
dict(type='Resize', scale=(1080, 736), keep_ratio=True),
|
|
# add loading annotation after ``Resize`` because ground truth
|
|
# does not need to do resize data transform
|
|
dict(
|
|
type='LoadOCRAnnotations',
|
|
with_polygon=True,
|
|
with_bbox=True,
|
|
with_label=True),
|
|
dict(
|
|
type='PackTextDetInputs',
|
|
meta_keys=('img_path', 'ori_shape', 'img_shape', 'scale_factor'))
|
|
]
|
|
|
|
ctw1500_textdet_train.pipeline = _base_.train_pipeline
|
|
ctw1500_textdet_test.pipeline = ctw_test_pipeline
|
|
|
|
train_dataloader = dict(
|
|
batch_size=8,
|
|
num_workers=4,
|
|
persistent_workers=True,
|
|
sampler=dict(type='DefaultSampler', shuffle=True),
|
|
dataset=ctw1500_textdet_train)
|
|
|
|
val_dataloader = dict(
|
|
batch_size=1,
|
|
num_workers=1,
|
|
persistent_workers=True,
|
|
sampler=dict(type='DefaultSampler', shuffle=False),
|
|
dataset=ctw1500_textdet_test)
|
|
|
|
test_dataloader = val_dataloader
|
|
|
|
auto_scale_lr = dict(base_batch_size=8)
|