New AI Model Reconstructs 3D Scenes in Real-Time at 20 FPS Across Thousands of Frames

A groundbreaking new AI model called LingBot-Map's Geometric Context Transformer reconstructs 3D scenes in real-time at 20 FPS across sequences exceeding 10,000 frames, offering a powerful open-source tool for streaming 3D scene reconstruction with interactive browser-based visualization.

• LingBot-Map is a feed-forward 3D foundation model called the Geometric Context Transformer, designed for real-time streaming 3D scene reconstruction using anchor context, pose-reference windows, and trajectory memory in a unified architecture.
• The model achieves high-efficiency streaming inference at approximately 20 FPS on 518×378 resolution over sequences exceeding 10,000 frames, utilizing paged KV cache attention via FlashInfer for stable long-sequence performance.
• Released under Apache License 2.0, the project offers multiple model checkpoints optimized for different sequence lengths, supports video and image input, includes sky masking for outdoor scenes, and provides an interactive browser-based 3D visualization tool.