ABot-Recon: Long-Horizon Streaming 3D Reconstruction with 12-Frames Context

zhenjun_zhao · x · 2026-08-31

The paper "Revisiting Local Context for Long-Horizon Streaming 3D Reconstruction" presents ABot-Recon, a streaming model that relies strictly on local context (caching KV features from only the preceding 11 frames). It predicts a point map in the current camera coordinate system and adjacent-frame relative poses. To reduce drift, it uses a lightweight temporal refiner and a composition-aware pose loss. Evaluations show superior long-horizon performance on challenging benchmarks like Oxford Spires.

Original post →

More from Research

Research channel →