| 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). |