Q-Instruct: Improving Low-level Visual Abilities for Multi-modality Foundation Models | Read Paper on Bytez