In the intermediate process state estimation method, two generators are used, and as well as inputting a common input noise to the respective generators, a label corresponding to a certain step is input to one generator and a label corresponding to a step different from the certain step is input to the other generator. Then, one of generation data and training data generated by the respective generators is randomly input to a discriminator, and the generators and the discriminator learn in an adversarial manner from the discrimination result in the discriminator. Then, an input noise corresponding to a desired final state and a label corresponding to a step where it is desired that an intermediate process state be estimated are input to the learned generator to estimate the intermediate process state based on the generation data generated by the generator.