Reproduce “LVMem: Image-Augmented Memory for Long-Horizon Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Victoria Khoruzhaya, Maksim Kalmykov, Tuong Do Minh, Tatiana Zemskova, Yixin Zhu, Dmitry Yudin
LVMem lets a robot search long video histories through descriptions linked to source images, using optical-flow frame selection and object-centered captions. When text retrieval is insufficient, the agent can inspect linked images. On OC-NaVQA with Qwen3.5-9B, it exceeds ReMEmbR in binary accuracy (76.12 ± 2.59% versus 65.17 ± 3.11%), text accuracy (39.39% versus 24.24%), and positional performance (50.12 versus 68.67 m); this comparison does not show gains on every metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.