We just released code for ROMAN (Robust Object Map Alignment Anywhere)! Check out the core code here and the ROS1/2 wrappers here.
ROMAN creates maps of open-set objects and uses their shape, semantic embeddings, and relative positions to robustly match objects between maps, enabling visual global localization even when robots are facing opposite directions. By successfully performing loop closure in challenging scenarios, ROMAN extends the capabilities of state-of-the-art visual SLAM. I hope it will be helpful/interesting to some users here!