Towards Addressing GAN Training Instabilities: Dual-objective GANs with Tunable Parameters