Reproduce “Xiaomi-Lens-0 Technical Report” 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
Pei Fu, Yike Liu, Yue Sun, Hanmo Wang, Xiuwen Xi, Ruoceng Zhang, Shaojie Zhang, Tingwei Huang and 8 more
Xiaomi-Lens-0 trains a multimodal model to connect language-model states for regions with pixel masks. A lightweight decoder turns each mask-token state and shared patch features into a mask; joint losses supervise both. It reaches 81.4 average cIoU, pooled pixel intersection-over-union, across eight referring-expression splits versus 78.6 for Qwen3-VL-Seg 4B, but does not lead on general LensBench images.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.