EVAThink: Evidence Sufficiency-Guided Training for Adaptive Thinking in Video Understanding

Researchers introduced EVAThink, a training method that uses evidence sufficiency to guide adaptive thinking in video-understanding vision-language models, according to the paper's abstract. EVAThink targets the trade-off under a fixed frame budget: when sampled frames supply sufficient evidence the model answers directly, and when observations are less informative it engages reasoning that connects them.

Under a fixed frame budget, video understanding with vision-language models VLMs depends on whether the sampled frames provide sufficient evidence for the question. Sufficient frame evidence supports direct answering, whereas less informative observations may benefit from reasoning that connects t