pyojinkim / facap

[IROS 2022]Floorplan-Aware Camera Poses Refinement

FACaP

This repository contains an implementation of Floorplan-Aware Camera Poses Refinement paper, which was presented on IROS 2022.

Floorplan-Aware Camera Poses Refinement
Anna Sokolova, Filipp Nikitin, Anna Vorontsova, Anton Konushin
Samsung AI Center Moscow
https://arxiv.org/abs/2210.04572

Introduction

For many indoor scenes, there exists an image of a technical floorplan that contains information about the geometry and main structural elements of the scene, such as walls, partitions, and doors. We argue that such a floorplan is a useful source of spatial information, which can guide a 3D model optimization.

The standard RGB-D 3D reconstruction pipeline consists of a tracking module applied to an RGB-D sequence and a bundle adjustment (BA) module that takes the posed RGB-D sequence and corrects the camera poses to improve consistency. We propose a novel optimization algorithm expanding conventional BA that leverages the prior knowledge about the scene structure in the form of a floorplan. Our experiments on the Redwood dataset and our self-captured data demonstrate that utilizing floorplan improves accuracy of 3D reconstructions.

Installation

We recommend using Dockerfile to build a container for the project. All required libraries are described in the environment file env.yaml. In the final version we drop all of the pytorch3d dependencies, so you do not have to install it.

Data structure

All scans should be preprocessed to the next structure:

scan
│   floorplan.npy
│   db.h5
│
└───arcore
│   │   cam_params-0001.txt
│   │   ...
│   │   depth-0001.png
│   │   ...
│   │   frame-0001.png
│   │   ...
│   │   pose-0001.txt
│   │   ...
│   │
│
└───segmentation
│   │   frame-0001_wall.png
│   │   ...
│   │   frame-0001_floor.png
│   │   ...
│  

Here:

• floorplan.npy is an array with the shape n x 4. Each element is a segment of the floorplan.
• db.h5 features a database in COLMAP format, which is used to map covisible points.
• cam_params-0001.txt intrinsics of the corresponding camera (w, h, f1, f1, p1, p2).
• pose-0001.txt extrinsic matrix of the corresponding camera
• depth-0001.png depth map
• frame-0001.png RGB frame
• frame-0001_wall.png rotated mask of walls for the corresponding frame
• frame-0001_floor.png rotated mask of the floor for the corresponding frame

For more details please see the file facap/data/scan.py.

Usage

To run an experiment you should create a config file run experiment.

    python scripts/run_experimnt.py --config path_to_config --device "cuda:0"

The example of the config can be found in the path experiments/config.yaml.

Citation

Citation

If you find this work useful for your research, please cite our paper:

@article{sokolova2022floorplan,
  title={Floorplan-Aware Camera Poses Refinement},
  author={Sokolova, Anna and Nikitin, Filipp and Vorontsova, Anna and Konushin, Anton},
  journal={arXiv preprint arXiv:2210.04572},
  year={2022}
}