Reproduce “MJev-Spatial: System One Spatial Intelligence via Native Attention Matching” 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
Luozheng Qin, Yuhan Zhu, Changhai Zhou, Qian Qiao, Jun Gao, Yuhua Zhou, Harry Zhang, Xiangyu Zeng and 5 more
MJev-Spatial answers spatial questions by selecting among candidate answers rather than generating text. It scores contextualized candidates with the pretrained query and key projections, handling choices, numerical intervals, and progressively refined image regions; point predictions require crop re-encoding. Across twelve benchmarks, it reports a 65.64% mean score versus 62.79% for Qwen3-VL-8B-Instruct, with results limited to these evaluations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.