[ACL2025 Findings] Migician: Revealing the Magic of Free-Form Multi-Image Grounding in Multimodal Large Language Models - View it on GitHub