session_init=get_model_loader(args.load),
input_names=['image'],
output_names=get_model_output_names()))
# autotune is too slow for inference
os.environ['TF_CUDNN_USE_AUTOTUNE'] = '0'
assert args.load
offline_pred([pred], args.evaluate)
else:
logger.set_logger_dir(args.logdir)
factor = get_batch_factor()
stepnum = config.STEP_PER_EPOCH
cfg = TrainConfig(
model=get_model(),
data=QueueInput(get_train_dataflow()),
callbacks=[
PeriodicCallback(ModelSaver(max_to_keep=10,
keep_checkpoint_every_n_hours=1),
every_k_epochs=20),
ScheduledHyperParamSetter('learning_rate',
[(40, config.BASE_LR * 0.1),
(60, config.BASE_LR * 0.01)]),
#EvalCallback(),
GPUUtilizationTracker(),
PeakMemoryTracker(),
EstimatedTimeLeft(),
],
steps_per_epoch=stepnum,
max_epoch=80,
session_init=get_model_loader(args.load) if args.load else None,
session_init=get_model_loader(args.load),
input_names=['image'],
output_names=get_model_output_names()))
# autotune is too slow for inference
os.environ['TF_CUDNN_USE_AUTOTUNE'] = '0'
assert args.load
offline_pred([pred], args.evaluate)
else:
logger.set_logger_dir(args.logdir)
factor = get_batch_factor()
stepnum = config.STEP_PER_EPOCH
cfg = TrainConfig(
model=get_model(),
data=QueueInput(get_train_dataflow()), #FeedInput QueueInput
callbacks=[
PeriodicCallback(ModelSaver(max_to_keep=10,
keep_checkpoint_every_n_hours=0.5),
every_k_epochs=1),
ScheduledHyperParamSetter(
'learning_rate',
[
(30, config.BASE_LR * 0.1), #
(60, config.BASE_LR * 0.01)
] #
),
# EvalCallback(),
# GPUUtilizationTracker(),
# PeakMemoryTracker(),
# EstimatedTimeLeft(),