lingbot-map
LibraryRobbyant/lingbot-map
A feed-forward 3D foundation model for streaming scene reconstruction.
Overview
LingBot-Map is a feed-forward 3D foundation model that reconstructs scenes from streaming data. It uses a Geometric Context Transformer to unify coordinate grounding, dense geometric cues, and drift correction. Achieves ~20 FPS on long sequences with a paged KV cache.
README Preview
\n \n\nLingBot-Map: Geometric Context Transformer for Streaming 3D Reconstruction\n\nRobbyant Team\n\n\n\n\n\n[](https://arxiv.org/abs/2604.14141)\n[](lingbot-map_paper.pdf)\n[](https://technology.robbyant.com/lingbot-map)\n[](https://huggingface.co/robbyant/lingbot-map)\n[](https://www.modelscope.cn/models/Robbyant/lingbot-map)\n[](LICENSE.txt)\n\n\n\nhttps://github.com/user-attachments/assets/fe39e095-af2c-4ec9-b68d-a8ba97e505ab\n\n-----\n\n### 🗺️ Meet LingBot-Map! We've built a feed-forward 3D foundation model for streaming 3D reconstruction! 🏗️🌍\n\nLingBot-Map has focused on:\n\n- **Geometric Context Transformer**: Architecturally unifies coordinate grounding, dense geometric cues, and long-range drift correction within a single streaming framework through anchor context, pose-reference window, and trajectory memory.\n- **High-Efficiency Streaming Inference**: A feed-forward architecture with paged KV cache attention, enabling stable inference at ~20 FPS on 518×378 resolution over long sequences exceeding 10,000 frames.\n- **State-of-the-Art Reconstruction**: Superior performance on diverse benchmarks compared to both existing streaming and iterative optimization-based approaches.\n\n---\n\n## 📑 Table of Contents\n\n\nClick to expand\n\n- [📰 News](#-news)\n- [📋 TODO](#-todo)\n- [⚙️ Installation](#️-installation)\n- [📦 Model Download](#-model-download)\n- [🚀 Quick Start](#-quick-start)\n- [🎬 Interactive Demo (`demo.py`)](#-interactive-demo-demopy)\n - [Try the Example Scenes](#try-the-example-scenes)\n - [Streaming with Keyframe Interval](#streaming-with-keyframe-interval)\n - [Windowed Inference (for long sequences, >3000 frames)](#windowed-inference-for-long-sequences-3000-frames)\n - [Sky Masking](#sky-masking)\n - [Visualization Options](#visualization-options)\n - [Performance & Memory](#performance--memory)\n- [🎥 Offline Rendering Pipeline (`demo_render/batch_demo.py`)](#-offline-rendering-pipeline-demo_renderbatch_demopy)\n- [📜 License](#
FAQ (2)
informationalWhy does the paper say 13 Oxford Spires scenes while the listed sequences total 14, and which sequences are actually evaluated?
The "13 scenes" description is a typo: the list contains 14 sequences. The authors initially evaluated all 14, obtaining ATE RMSE = 5.59, but found that poses and point clouds were misaligned in 4 scenes. The final paper therefore reports results on 10 scenes, with ATE RMSE = 6.42.
Excluded sequences:
- blenheim-palace-01: 4.116
- blenheim-palace-02: 3.687
- blenheim-palace-05: 2.567
- christ-church-01: 3.6793
Included sequences and available per-sequence ATE values (first 320 frames, sampled every 12 frames):
- bodleian-library-02: 9.592
- christ-church-02: 3.2969
- christ-church-03: 0.4530
- christ-church-05: 35.987
- keble-college-02: 3.016
- keble-college-03: 1.957
- keble-college-04, keble-college-05, observatory-quarter-01, and observatory-quarter-02 are included, but their individual values were cut off in the discussion thread.
The authors stated that the scene description will be corrected in the next paper version.
TroubleshootingWhich checkpoint is used to reproduce the Oxford Spires Table 2 results?
Use the **160th-epoch checkpoint from lingbot-map.pt** for the Table 2 Oxford Spires result. The maintainers stated that both lingbot-map.pt and lingbot-map-long.pt can achieve better ATE than the reported Table 2 value, but lingbot-map-long.pt is not the checkpoint used for that table. For AUC, follow the **DA3 pose-AUC calculation**. The exact image preprocessing, crop/resize, intrinsics handling, and official evaluation code were not provided in the thread; the maintainers said the evaluation code would be released, so use that official code to resolve remaining AUC gaps.