assign requires shapes of both tensors to matchの解決法

概要

以下のようなエラーが出た際の対処法

restoring from checkpoint failed. this is most likely due to a mismatch between 
the current graph and the graph from the checkpoint. please ensure that you have 
not altered the graph expected based on the checkpoint

内容

以下のようにtensoflow1系でのsavedmodelを二つ読み込む際に一度runしたモデルでもう一方もrunしてしまう

with tf.Session() as sess:
    model = tf.saved_model.loader.load(sess, [tag_constants.SERVING], export_path1)
    sig = model_x.signature_def[signature_name]
    inp = sig_def.inputs[input].name
    out = sig_def.outputs[output].name
    output = sess.run(out,
                 feed_dict={inp: my_data})

with tf.Session() as sess:
    model = tf.saved_model.loader.load(sess, [tag_constants.SERVING], export_path2)
    sig = model_x.signature_def[signature_name]
    inp = sig_def.inputs[input].name
    out = sig_def.outputs[output].name
    output = sess.run(out,
                 feed_dict={inp: my_data})

解決法

Session()内にgraph=tf.Graph()を記載する

with tf.Session(graph=tf.Graph()) as sess: