On the transferability of adversarial examples between convex and 01 loss models