README.md

---
license: apache-2.0
pipeline_tag: image-to-3d
---

# ABot-Recon

ABot-Recon is a streaming 3D reconstruction model that estimates camera motion and scene geometry online from extremely long videos using only a fixed local context of 12 frames. It predicts a point map in the current camera coordinate system and an adjacent-frame relative pose, then composes these local predictions into a global reconstruction through sequential composition.

**Paper:** [Revisiting Local Context for Long-Horizon Streaming 3D Reconstruction](https://arxiv.org/abs/2608.27529)  
**Project page:** [ABot-Recon](https://amap-cvlab.github.io/ABot-Recon-html/)  
**Code:** [github.com/amap-cvlab/ABot-Recon](https://github.com/amap-cvlab/ABot-Recon)

## Quick Start

```python
from pathlib import Path
from abot_recon import ABotRecon

images = sorted(Path("examples/images").glob("*.jpg"))

model = ABotRecon.from_pretrained(
    "acvlab/ABot-Recon",
    device="cuda",
    attention_backend="auto",
    loop_closure=False,
)

result = model.infer(images)

trajectory = result.camera_poses
relative_poses = result.relative_poses
local_points = result.local_points
confidence = result.confidence
```

For a full description of usage options, please refer to the [GitHub README](https://github.com/amap-cvlab/ABot-Recon).