检查输入时Keras形状错误

我正在尝试训练一个简单的MLP模型，该模型将输入问题（使用300D字嵌入）和使用预训练VGG16模型提取的图像特征映射到固定长度的特征向量 . 但是，我无法弄清楚如何解决下面提到的错误 . 这是我正在尝试运行的代码：

parser = argparse.ArgumentParser()
parser.add_argument('-num_hidden_units', type=int, default=1024)
parser.add_argument('-num_hidden_layers', type=int, default=3)
parser.add_argument('-dropout', type=float, default=0.5)
parser.add_argument('-activation', type=str, default='tanh')
parser.add_argument('-language_only', type=bool, default= False)
parser.add_argument('-num_epochs', type=int, default=10) #default=100
parser.add_argument('-model_save_interval', type=int, default=10)
parser.add_argument('-batch_size', type=int, default=128)
args = parser.parse_args()

questions_train = open('data/qa/preprocess/questions_train2014.txt', 'r').read().splitlines()
answers_train = open('data/qa/preprocess/answers_train2014_modal.txt', 'r').read().splitlines()
images_train = open('data/qa/preprocess/images_train2014.txt', 'r').read().splitlines()
vgg_model_path = 'data/coco/vgg_feats.mat'
maxAnswers = 1000
questions_train, answers_train, images_train = selectFrequentAnswers(questions_train,answers_train,images_train, maxAnswers)

#encode the remaining answers
labelencoder = preprocessing.LabelEncoder()
labelencoder.fit(answers_train)
nb_classes = len(list(labelencoder.classes_))
joblib.dump(labelencoder,'models/labelencoder.pkl')

features_struct = scipy.io.loadmat(vgg_model_path)
VGGfeatures = features_struct['feats']
print ('loaded vgg features')
image_ids = open('data/coco/coco_vgg_IDMap.txt').read().splitlines()
id_map = {}
for ids in image_ids:
    id_split = ids.split()
    id_map[id_split[0]] = int(id_split[1])

nlp = English()
print ('loaded word2vec features...')
img_dim = 4096
word_vec_dim = 300

model = Sequential()
if args.language_only:
    model.add(Dense(args.num_hidden_units, input_dim=word_vec_dim, init='uniform'))
else:
    model.add(Dense(args.num_hidden_units, input_dim=img_dim+word_vec_dim, init='uniform'))
model.add(Activation(args.activation))
if args.dropout>0:
    model.add(Dropout(args.dropout))
for i in range(args.num_hidden_layers-1):
    model.add(Dense(args.num_hidden_units, init='uniform'))
    model.add(Activation(args.activation))
    if args.dropout>0:
        model.add(Dropout(args.dropout))
model.add(Dense(nb_classes, init='uniform'))
model.add(Activation('softmax'))

json_string = model.to_json()
if args.language_only:
    model_file_name = 'models/mlp_language_only_num_hidden_units_' + str(args.num_hidden_units) + '_num_hidden_layers_' + str(args.num_hidden_layers)
else:
    model_file_name = 'models/mlp_num_hidden_units_' + str(args.num_hidden_units) + '_num_hidden_layers_' + str(args.num_hidden_layers)     
open(model_file_name  + '.json', 'w').write(json_string)

print ('Compiling model...')
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')
print ('Compilation done...')

print ('Training started...')
for k in range(args.num_epochs):
    #shuffle the data points before going through them
    index_shuf = list(range(len(questions_train)))
    shuffle(index_shuf)
    questions_train = [questions_train[i] for i in index_shuf]
    answers_train = [answers_train[i] for i in index_shuf]
    images_train = [images_train[i] for i in index_shuf]
    progbar = generic_utils.Progbar(len(questions_train))
    for qu_batch,an_batch,im_batch in zip(grouper(questions_train, args.batch_size, fillvalue=questions_train[-1]), 
                                        grouper(answers_train, args.batch_size, fillvalue=answers_train[-1]), 
                                        grouper(images_train, args.batch_size, fillvalue=images_train[-1])):
        X_q_batch = get_questions_matrix_sum(qu_batch, nlp)
        if args.language_only:
            X_batch = X_q_batch
        else:
            X_i_batch = get_images_matrix(im_batch, id_map, VGGfeatures)
            X_batch = np.hstack((X_q_batch, X_i_batch))
        Y_batch = get_answers_matrix(an_batch, labelencoder)
        loss = model.train_on_batch(X_batch, Y_batch)            
        progbar.add(args.batch_size, values=[("train loss", loss)])
    #print type(loss)
    if k%args.model_save_interval == 0:
        model.save_weights(model_file_name + '_epoch_{:02d}.hdf5'.format(k))

model.save_weights(model_file_name + '_epoch_{:02d}.hdf5'.format(k))

这是我得到的错误：

Keras：检查输入时出错：期望dense_9_input有形状（4396，）但是有形状的数组（4096，）