@YinghaoXu1
š After one year of teamwork, we are excited to release our 3D foundation model ā LingBot-Map! Unlike DA3/VGGT, LingBot-Map is a purely autoregressive model for streaming 3D reconstruction ā” It achieves ~20 FPS on 518Ć378 resolution over sequences exceeding 10,000 frames ā and beyond š Two key insights behind LingBot-Map: š Keep SLAM's structural wisdom: build Geometric Context Attention with long-context modeling while maintaining a compact streaming state š Make everything end-to-end learnable ā no optimization, no post-processing Let's check out our demos š