Tag
Introduces DMV-Bench, an interactive benchmark for evaluating visual memory in multimodal agents using incidental visual cues from product images, and proposes DualMem, a dual-coding memory architecture that outperforms text-only and other multimodal baselines across various chain lengths.