File size: 8,017 Bytes
d670799 |
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 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 |
# Copyright (c) OpenMMLab. All rights reserved.
import argparse
from mmengine.config import Config
def convert(config_path, output_config_path):
print('start convert')
cfg = Config.fromfile(config_path)
origin_dataset_type = cfg.dataset_type
# dataset
if origin_dataset_type != 'VideoDataset':
cfg.dataset_type = 'VideoDataset'
cfg.data_root = 'data/kinetics400/rawframes_train'
cfg.data_root_val = 'data/kinetics400/rawframes_val'
cfg.ann_file_train = \
'data/kinetics400/kinetics400_train_list_rawframes.txt'
cfg.ann_file_val = \
'data/kinetics400/kinetics400_val_list_rawframes.txt'
cfg.ann_file_test = \
'data/kinetics400/kinetics400_val_list_rawframes.txt'
# model
preprocess_cfg = cfg.img_norm_cfg
formatshape = None
for trans in cfg.train_pipeline:
if trans.type == 'FormatShape':
formatshape = trans.input_format
preprocess_cfg['input_format'] = formatshape
cfg.preprocess_cfg = preprocess_cfg
cfg.model.data_preprocessor = dict(
type='ActionDataPreprocessor', **dict(preprocess_cfg))
cfg.pop('img_norm_cfg')
if (cfg.model.test_cfg is not None) and ('average_clips'
in cfg.model.test_cfg):
cfg.model.cls_head.average_clips = cfg.model.test_cfg.average_clips
cfg.model.test_cfg.pop('average_clips')
if len(cfg.model.test_cfg) == 0:
cfg.model.test_cfg = None
# pipeline
pipelines = [cfg.train_pipeline, cfg.val_pipeline, cfg.test_pipeline]
if origin_dataset_type == 'VideoDataset':
for pipeline in pipelines:
new_pipeline = [
trans for trans in pipeline
if trans.type not in ['Normalize', 'Collect', 'ToTensor']
]
new_pipeline.append(dict(type='PackActionInputs'))
pipeline.clear().extend(new_pipeline)
elif origin_dataset_type == 'RawframeDataset':
for pipeline in pipelines:
new_pipeline = [
trans for trans in pipeline if trans.type not in
['RawFrameDecode', 'Normalize', 'Collect', 'ToTensor']
]
new_pipeline.insert(0, dict(type='DecordInit'))
new_pipeline.insert(2, dict(type='DecordDecode'))
new_pipeline.append(dict(type='PackActionInputs'))
pipeline.clear()
pipeline.extend(new_pipeline)
# dataloader
cfg.data.train.update(
dict(
type=cfg.dataset_type,
ann_file=cfg.ann_file_train,
data_prefix=dict(video=cfg.data_root),
pipeline=cfg.train_pipeline))
cfg.train_dataloader = dict(
batch_size=cfg.data.videos_per_gpu,
num_workers=cfg.data.workers_per_gpu,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=cfg.data.train)
val_batchsize = cfg.data.val_dataloader.videos_per_gpu \
if 'val_dataloader' in cfg.data else cfg.data.videos_per_gpu
cfg.data.val.update(
dict(
type=cfg.dataset_type,
ann_file=cfg.ann_file_val,
data_prefix=dict(video=cfg.data_root_val),
pipeline=cfg.val_pipeline))
cfg.val_dataloader = dict(
batch_size=val_batchsize,
num_workers=cfg.data.workers_per_gpu,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=cfg.data.val)
test_batchsize = cfg.data.test_dataloader.videos_per_gpu \
if 'test_dataloader' in cfg.data else cfg.data.videos_per_gpu
cfg.data.test.update(
dict(
type=cfg.dataset_type,
ann_file=cfg.ann_file_test,
data_prefix=dict(video=cfg.data_root_val),
pipeline=cfg.test_pipeline))
cfg.test_dataloader = dict(
batch_size=test_batchsize,
num_workers=cfg.data.workers_per_gpu,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=cfg.data.test)
cfg.pop('data')
# eval
cfg.val_evaluator = dict(type='AccMetric')
cfg.test_evaluator = cfg.val_evaluator
cfg.val_cfg = dict(interval=cfg.evaluation.interval)
cfg.test_cfg = dict()
cfg.pop('evaluation')
# optimizer
optimizer_wrapper = dict(optimizer=dict())
for k, v in cfg.optimizer.items():
if k not in ['paramwise_cfg', 'constructor']:
optimizer_wrapper['optimizer'].update({k: v})
else:
optimizer_wrapper.update({k: v})
for k, v in cfg.optimizer_config.items():
if k == 'grad_clip':
k = 'clip_grad'
optimizer_wrapper.update({k: v})
cfg.optimizer_wrapper = optimizer_wrapper
cfg.pop('optimizer')
cfg.pop('optimizer_config')
# train_cfg
cfg.train_cfg = dict(by_epoch=True, max_epochs=cfg.total_epochs)
cfg.pop('total_epochs')
# schedule
cfg.param_scheduler = []
warmup_epoch = 0
if 'warmup' in cfg.lr_config:
warmup_ratio = 0.1 \
if 'warmup_ratio' not in cfg.lr_config \
else cfg.lr_config.warmup_ratio
warmup_epoch = cfg.lr_config.warmup_iters
cfg.param_scheduler.append(
dict(
type='LinearLR',
bengin=0,
start_factor=warmup_ratio,
end=cfg.lr_config.warmup_iters,
by_epoch=cfg.lr_config.warmup_by_epoch))
if cfg.lr_config.policy == 'step':
cfg.param_scheduler.append(
dict(
type='MultiStepLR',
milestones=cfg.lr_config.step,
by_epoch=cfg.train_cfg.by_epoch,
begin=0,
end=cfg.train_cfg.max_epochs,
gamma=0.1
if 'gamma' not in cfg.lr_config else cfg.lr_config.gamma))
elif cfg.lr_config.policy == 'CosineAnnealing':
cfg.param_scheduler.append(
dict(
type='CosineAnnealingLR',
eta_min=cfg.lr_config.min_lr,
by_epoch=cfg.train_cfg.by_epoch,
begin=warmup_epoch,
end=cfg.train_cfg.max_epochs,
T_max=cfg.train_cfg.max_epochs - warmup_epoch))
else:
raise ValueError(f'Not support convert {cfg.lr_config.policy}')
cfg.pop('lr_config')
# runtime
cfg.default_scope = 'mmaction'
cfg.default_hooks = dict(
runtime_info=dict(type='RuntimeInfoHook'),
timer=dict(type='IterTimerHook'),
logger=dict(type='LoggerHook', interval=20),
param_scheduler=dict(type='ParamSchedulerHook'),
checkpoint=dict(type='CheckpointHook', **cfg.checkpoint_config),
sampler_seed=dict(type='DistSamplerSeedHook'),
)
cfg.env_cfg = dict(
cudnn_benchmark=False,
mp_cfg=dict(
mp_start_method=cfg.mp_start_method,
opencv_num_threads=cfg.opencv_num_threads),
dist_cfg=dict(**cfg.dist_params),
)
cfg.log_level = 'INFO'
cfg.load_from = None
cfg.resume = False
cfg.pop('workflow')
cfg.pop('mp_start_method')
cfg.pop('opencv_num_threads')
cfg.pop('log_config')
cfg.pop('dist_params')
cfg.pop('checkpoint_config')
cfg.pop('work_dir')
cfg.dump(output_config_path)
print('Successful')
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='Convert an action recognizer config file \
from OpenMMLAb framework v1.0 to v2.0')
parser.add_argument('config', help='The config file path')
parser.add_argument('output_config', help='The config file path')
args = parser.parse_args()
convert(args.config, args.output_config)
|