Few-Shot Image-to-Semantics Translation for Policy Transfer in Reinforcement Learning