Different output between .pb and .ckpt file

[YIYANGCAI]

Hello! Thanks for this great work of text detection. I am recently working on a task of text detection and applying this model. the performance on my own data after training is wonderful. However, when I converted my training model of ckpt to pb model, the inference result of pb changes a lot. I cannot find out where the problem is, could you help me to locate my problem?

here is my conversion code from ckpt to pb model:

def freezeGraph(input_checkpoint, output_nodes_names, output_graph):
    '''
    :param input_checkpoint:
    :param output_graph: 
    :return:
    '''
    # checkpoint = tf.train.get_checkpoint_state(model_folder) 
    # input_checkpoint = checkpoint.model_checkpoint_path 

    # 
    output_node_names = output_nodes_names
    saver = tf.train.import_meta_graph(input_checkpoint + '.meta', clear_devices=True)
    graph = tf.get_default_graph() # 
    input_graph_def = graph.as_graph_def()  # 

    with tf.Session() as sess:
        saver.restore(sess, input_checkpoint) #
        output_graph_def = graph_util.convert_variables_to_constants(  # 
            sess=sess,
            input_graph_def=input_graph_def,# 
            output_node_names=output_node_names)# 

        with tf.gfile.GFile(output_graph, "wb") as f: 
            f.write(output_graph_def.SerializeToString()) 
        print("%d ops in the final graph." % len(output_graph_def.node)) 

if __name__ == '__main__':
    nodes = getOutNodes('./seglink.txt') 
    freezeGraph('./ckpt/model.ckpt-5882', nodes, './pb/seglink.pb')

the seglink.txt is recording the nodes' name of nodes seglink.txt

I wonder if its the problem of conversion

With many thanks if you can offer me some help of this problem!