cifar10在resnet18上出现了问题Tensorflow

因为比赛需要用到ResNet，要自己写一个网络框架，就自己手写了一个ResNet18。使用的数据集是cifar10，最后跑出来的结果很差，只有50%左右，并且震荡严重（有时候准确率会下降）。有大神能帮忙指出下问题在哪吗？感谢！
以下是我手写的resnet18代码，cifar数据集是手动下载的。

import numpy as np
import math
import cv2 as cv
import pickle
import tensorflow as tf
import matplotlib.pyplot as plt
import os

# os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
# mnist = input_data.read_data_sets('MNIST_data',one_hot = True)
#参数配置
train_step = 2000
batch_size = 32

def load_file(filename):
    with open(filename, 'rb') as fo:
        data = pickle.load(fo,encoding = 'latin1')
    return data

class Cifar:
    def __init__(self, filename, need_shuffle):
        all_data = []
        all_label = []
        for each in filename:
            temp = load_file(each)
            data = temp['data']
            all_data.append(data)
            label = temp['labels']
            all_label.append(label)
        self._data = np.vstack(all_data)
        self._data = (self._data /127.5)-1
        self._label = np.hstack(all_label)
        self._example_num = self._data.shape[0]
        self._shuffle = need_shuffle
        self._begin = 0
        if self._shuffle:
            self._shuffle_data()
    def _shuffle_data(self):
        p = np.random.permutation(self._example_num)
        self._data = self._data[p]
        self._label = self._label[p]
    def next_batch(self, batch_size):
        end = self._begin + batch_size
        if end > self._example_num:
            if self._shuffle:
                self._shuffle_data()
                self._begin = 0
                end = batch_size
            else:
                raise Exception('error1')
        if end>self._example_num:
            raise Exception('error2')
        out_data = self._data[self._begin: end]
        out_label = self._label[self._begin: end]
        self._begin = end
        return out_data,out_label

def Resnet50(input,train_):
    #第0层卷积
    with tf.variable_scope('conv0_'):
        w0 = tf.get_variable('conv0', [5,5,3,64], initializer = tf.truncated_normal_initializer(stddev = 0.1), dtype = tf.float16)
        y0 = tf.nn.conv2d(input, w0, strides = [1,2,2,1], p