efficientdet 실습 질문

class TRAIN_CFG:
    model_name='efficientdet-d2'
    strategy=''
    model_dir=model_weight
    pretrained_ckpt = ckpt_path#coco로 pretrained된 checkpoint파일이 있는 디렉토리 위치

    #num_classes=3으로 변경
    hparams='num_classes=3,moving_average_decay=0,mixed_precision=true'
    use_xla=False
    use_fake_data=False

    #max_instances_per_image를 200으로 설정하고
    #d2모델이 상대적으로 더 크기 때문에 batch_size를 8로 설정시 GPU Out of Memory발생
    batch_size=4
    eval_samples=5000 #evaluation image데이터 갯수
    steps_per_execution=1 #train시 steps 횟수
    num_examples_per_epoch=3000 #1epochs 시 적용하는 examples 개수
    num_epochs=20 #epochs 횟수
    train_file_pattern=train_temp+'/*.tfrecord'
    test_file_pattern=test_temp+'/*.tfrecord'
    test_json_file=None #optional coco validation json
    mode='traineval' #train만 적용 또는 train과 eval함께 적용(traineval)
    num_shards=100
    max_instances_per_image=100

    num_cores=2
    tpu=None
    gcp_project=None
    tpu_zone=None
    eval_master=''
    eval_name=None
    tf_random_seed=2021
    profile=False
    debug=False

sys.path.append('./automl/efficientdet')

from keras.train import setup_model
import hparams_config

import utils
from keras import tfmot
from keras import train_lib
from keras import util_keras

config=hparams_config.get_detection_config(TRAIN_CFG.model_name)
config.override(TRAIN_CFG.hparams)

steps_per_epoch=TRAIN_CFG.num_examples_per_epoch // TRAIN_CFG.batch_size

if tf.config.list_physical_devices('GPU'):
    ds_strategy=tf.distribute.OneDeviceStrategy('device:GPU:0')
else:
    ds_strategy=tf.distribute.OneDeviceStrategy('device:CPU:0')

print(ds_strategy)


params=dict(
    profile=TRAIN_CFG.profile,
    mode=TRAIN_CFG.mode,
    model_name=TRAIN_CFG.model_name,
    steps_per_execution=TRAIN_CFG.steps_per_execution,
    num_epochs=TRAIN_CFG.num_epochs,
    model_dir=TRAIN_CFG.model_dir,
    steps_per_epoch=steps_per_epoch,
    strategy=TRAIN_CFG.strategy,
    batch_size=TRAIN_CFG.batch_size,
    tf_random_seed=TRAIN_CFG.tf_random_seed,
    debug=TRAIN_CFG.debug,
    test_json_file=TRAIN_CFG.test_json_file,
    eval_samples=TRAIN_CFG.eval_samples,
    num_shards=ds_strategy.num_replicas_in_sync,
    max_instances_per_image = TRAIN_CFG.max_instances_per_image
)

config.override(params, True)

#image size를 tuple 형태로 변환. 512는 (512,512)로 '1920x880'은 (1920, 880)으로 변환
config.image_size=utils.parse_image_size(config.image_size)
print(config)

sys.path.append('./automl/efficientdet')

from keras import train
import numpy as np

config.batch_size=4
train_steps_per_epoch=train_df.shape[0]
test_steps_per_epoch=test_df.shape[0]
print('train_steps_per_epoch : ',train_steps_per_epoch,
    'test_steps_per_epoch : ', test_steps_per_epoch)

test_dataset=get_dataset(False, config) if 'eval' in config.mode else None
model.fit(
    get_dataset(True, config),
    epochs=20,
    steps_per_epoch=train_steps_per_epoch,
    callbacks=train_lib.get_callbacks(config.as_dict(), test_dataset),
    validation_data=test_dataset,
    validation_steps=test_steps_per_epoch
)

tf.keras.backend.clear_session()