Robbyant/lingbot-map +2,065 ★ this week + Follow

What it actually is

LingBot-Map is a feed-forward 3D foundation model for streaming 3D reconstruction, built around a Geometric Context Transformer (GCT). The core model code lives in lingbot_map/models/, with the forward pass defined in gct_base.py (forward, called by 16) and streaming inference logic in gct_stream.py (clean_kv_cache and _set_skip_append, each called by 14 and 10). It targets researchers and engineers working on SLAM, visual reconstruction, and 3D perception who need to process long image/video sequences into geometry and camera poses.

Why it's growing

The 2,065 stars this week coincide with an arXiv paper release, published model weights on both HuggingFace and ModelScope, and headline claims of ~20 FPS inference at 518×378 resolution over sequences exceeding 10,000 frames. The README frames the work as state-of-the-art versus both streaming and optimisation-based approaches, and the News section shows steady, dated feature releases (benchmark, long-video demos, acceleration). Without release notes or commit data fetched, the growth is best explained by the paper/model launch combined with concrete performance claims rather than any single measurable event in the graph facts.