Code for our work: Sa2VA: Marrying SAM2 with LLaVA for Dense Grounded Understanding of Images and Videos - View it on GitHub