LLM2R: From Large Language Models to 3D Reassemblers

One-line takeaway

NUS's LLM2R casts 3D part assembly as sequence modeling with a pretrained decoder-only LLM: a Point-BERT encoder and Q-Former compress each part into geometric tokens, and all parts are processed together to predict each part's rigid pose, scaling reasoning to objects with many parts.

新加坡国立大学的 LLM2R 把 3D 零件装配当作序列建模，用预训练的仅解码器大模型来做：Point-BERT 编码器加 Q-Former 把每个零件压成一组几何 token，所有零件一起输入，预测每个零件的刚体位姿，以应对零件很多时的整体推理。