防止keras占满GPU显存-keras-yolo

import os
import numpy as np
import keras.backend as K
from keras.layers import Input, Lambda
from keras.models import Model
from keras.optimizers import Adam
from keras.callbacks import TensorBoard, ModelCheckpoint, ReduceLROnPlateau, EarlyStopping
# 指定那几块GPU是对于本程序可见的 其实有没有都没有什么区别 tensor默认占用全部的
# 一般用在忽略哪块GPU 多GPU只能用一块那种任务
from yolo3.model import preprocess_true_boxes, yolo_body, tiny_yolo_body, yolo_loss
from yolo3.utils import get_random_data
# 多Gpu训练
from keras.utils import multi_gpu_model
import keras.backend.tensorflow_backend as KTF
import tensorflow as tf

os.environ["CUDA_VISIBLE_DEVICES"] = "1"

def _main():
    # 训练前记得修改yolov3.cfg和model_data/voc_class.txt
    annotation_path = '1.14/train/label1.txt' # 这里是要的由VOC生成的txt，里面包括image地址和框和种类 需修改
    log_dir = '1.14/weights/' # 是存放生成好的新的weights的
    classes_path = '1.14/train/classes.txt' # 这个我只标记了4类，所以要修改，原来voc是有20类的
    anchors_path = '1.14/train/anchors.txt' # 这里的anchor暂时先不修改(最好从新kmean算anchors)
    class_names = get_classes(classes_path)
    num_classes = len(class_names)
    anchors = get_anchors(anchors_path)

    # 修改大小，按照图片最小的边来修改
    input_shape = (544,960) # multiple of 32, hw 这里是否需要修改

    config = tf.ConfigProto()
    config.gpu_options.allow_growth=True   #不全部占满显存, 按需分配
    sess = tf.Session(config=config)

    KTF.set_session(sess)

    is_tiny_version = len(anchors)==6 # default setting
    if is_tiny_version:
        model = create_tiny_model(input_shape, anchors, num_classes,
            freeze_body=2, weights_path='model_data/tiny_yolo_weights.h5')
    else:
        # 这里要修改，这里的权重是基础权重，在这个基础上继续训练
        model = create_model(input_shape, anchors, num_classes,
            freeze_body=2, weights_path='12.5/weigths/dog_detect_544x960.h5') # make sure you know what you freeze