SECOND: Mitigating Perceptual Hallucination in Vision-Language Models via Selective and Contrastive Decoding | Read Paper on Bytez