Abstract: Training a deep learning inversion network usually requires hundreds of thousands of complex velocity models, which is labor-intensive and expensive to acquire. In this work, we develop a ...