About LLaVA-Next-Qwen2 UniRes Result.

[hcwei13]

This work is fantastic and has been very inspiring to me.

I have a questions: LLaVA-Next-Qwen2 supports a context length of 128K, roughly supporting 888 frames. However, Table 4 in the paper only shows results for 32 frames (AnyRes). Have you conducted more experiments using LLaVA-Next-Qwen2+UniRes on more frames, such as 256? This might provide a fairer comparison.

[jzhang38]

We have a heatmap for LLaVA-Next-Qwen2 on V-NIAH here. We will update our paper soon.