Tag
PSRD framework halves multimodal hallucination in LVLMs by using phase-wise self-reward decoding and a distilled lightweight reward model without extra supervision.