Datasets and models in the paper SSL4RL: Revisiting Self-supervised Learning as Intrinsic Reward for Visual-Language Reasoning