ctc 验证码识别 ValueError: No gradients provided for any variable

Question

from captcha.image import ImageCaptcha
import os
import string
import tensorflow as tf
from tensorflow import keras
import numpy as np
import matplotlib.pyplot as plt
import cv2
import random
chars = string.digits # 验证码字符集
nclass=len(chars)+1
print(nclass)
#生成验证码
def generate_img(img_dir: '图片保存目录'='img_dir'):
    for length in range(4, 8): # 验证码长度
        if not os.path.exists(f'{img_dir}/{length}'):
            os.makedirs(f'{img_dir}/{length}')
        for _ in range(1000):
            img_generator = ImageCaptcha()
            char = ''.join([random.choice(chars) for _ in range(length)])
            img_generator.write(chars=char, output=f'{img_dir}/{length}/{char}_{_}.jpg')
# generate_img("./image")
char_map = {chars[c]: c for c in range(len(chars))} # 验证码编码（0到len(chars) - 1)

#处理图片
def load_img(img_dir: '图片保存目录'='img_dir', min_length: '最小长度'=4, max_length: '最大长度'=7):
    labels = {length: [] for length in range(min_length, max_length + 1)} # 验证码真实标签{长度：标签列表}
    imgs = {length: [] for length in range(min_length, max_length + 1)} # 图片BGR数据字典{长度：BGR数据列表}
    ### 读取图片
    for length in range(min_length, max_length + 1):
        for file in os.listdir(f'{img_dir}/{length}'):
            img = cv2.imread(f'{img_dir}/{length}/{file}')
            labels[length].append(file[:-4].split('_')[0])
            height, width, _ = img.shape
            h_resize = 32
            w_resize = int(img.shape[1] * h_resize / img.shape[0])
            img_gray = cv2.cvtColor(cv2.resize(img, (w_resize, h_resize)), cv2.COLOR_BGR2GRAY) # 缩小图片固定宽度为32，并转为灰度图
            imgs[length].append(img_gray)

    ### 编码真实标签
    labels_encode = {length: [] for length in range(min_length, max_length + 1)}
    for length in range(min_length, max_length + 1):
        for label in labels[length]:
            label = [char_map[i] for i in label]
            labels_encode[length].append(label)
    return imgs, labels,labels_encode

def generate_data(imgs, labels_encode,length ):
    imgs = {length: np.array(imgs[length]) for length in range(4, 8)} # 图片BGR数据字典{长度：BGR数据数组}
    labels_encode = {length: np.array(labels_encode[length]) for length in range(4, 8)} # 验证码真实标签{长度：标签数组}
    x=imgs[length]
    x=x.reshape(-1,32,85,1)
    y=labels_encode[length]
    return [x, y,np.ones(len(x)) * int(42),np.ones(len(x)) * int(len(y[0]))],np.ones(len(x))


def ctc_lambda_func(args):
    y_pred, labels, input_length, label_length = args
#     y_pred = y_pred[:, 2:, :]
    return keras.backend.ctc_batch_cost(labels, y_pred, input_length, label_length)

input_tensor=keras.layers.Input(shape=(32,None,1))

x=keras.layers.Conv2D(32,kernel_size=3, padding='same', kernel_initializer='he_uniform',activation='relu')(input_tensor)
x=keras.layers.BatchNormalization()(x)
x=keras.layers.MaxPool2D((2,2))(x)

x=keras.layers.Conv2D(64,kernel_size=3, padding='same', kernel_initializer='he_uniform',activation='relu')(x)
x=keras.layers.BatchNormalization()(x)
x=keras.layers.Conv2D(64,kernel_size=3, padding='same', kernel_initializer='he_uniform',activation='relu')(x)

x = keras.layers.Permute((2, 1, 3))(x)
x =keras.layers.TimeDistributed(keras.layers.Flatten())(x)
x = keras.layers.Bidirectional(keras.layers.GRU(128, return_sequences=True,reset_after=False))(x)
x = keras.layers.Bidirectional(keras.layers.GRU(128, return_sequences=True,reset_after=False))(x)
x = keras.layers.Dense(nclass, activation='softmax')(x)

base_model = keras.Model(inputs=input_tensor, outputs=x)

labels=keras.layers.Input(shape=[None,],dtype=tf.float32,name='labels')
input_length=keras.layers.Input(shape=[1],dtype=tf.int64,name='input_length')
labels_length=keras.layers.Input(shape=[1],dtype=tf.int64,name='labels_length')
loss_out=keras.layers.Lambda(ctc_lambda_func,output_shape=(1,),name='ctc')([x,labels,input_length,labels_length])

model=keras.Model(inputs=[input_tensor,labels,input_length,labels_length],outputs=loss_out)


# 模型优化器
sgd = keras.optimizers.SGD(lr=0.0001, decay=1e-6, momentum=0.9, nesterov=True, clipnorm=5)

model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=sgd)

model.summary()

imgs, labels,labels_encode=load_img("./image/",4,7)
# for i in range(4,8):
model.fit_generator(generate_data(imgs, labels_encode, 4),epochs=10)

出现这样的错误：ValueError: No gradients provided for any variable: ['conv2d/kernel:0', 'conv2d/bias:0', 'batch_normalization/gamma:0', 'batch_normalization/beta:0', 'conv2d_1/kernel:0', 'conv2d_1/bias:0', 'batch_normalization_1/gamma:0', 'batch_normalization_1/beta:0', 'conv2d_2/kernel:0', 'conv2d_2/bias:0', 'bidirectional/forward_gru/gru_cell_1/kernel:0', 'bidirectional/forward_gru/gru_cell_1/recurrent_kernel:0', 'bidirectional/forward_gru/gru_cell_1/bias:0', 'bidirectional/backward_gru/gru_cell_2/kernel:0', 'bidirectional/backward_gru/gru_cell_2/recurrent_kernel:0', 'bidirectional/backward_gru/gru_cell_2/bias:0', 'bidirectional_1/forward_gru_1/gru_cell_4/kernel:0', 'bidirectional_1/forward_gru_1/gru_cell_4/recurrent_kernel:0', 'bidirectional_1/forward_gru_1/gru_cell_4/bias:0', 'bidirectional_1/backward_gru_1/gru_cell_5/kernel:0', 'bidirectional_1/backward_gru_1/gru_cell_5/recurrent_kernel:0', 'bidirectional_1/backward_gru_1/gru_cell_5/bias:0', 'dense/kernel:0', 'dense/bias:0'].

求大神帮忙解决！！！！

Answer 1

https://blog.csdn.net/qq_35276242/article/details/97172432