This repository contains the code for paper "An Indoor Scene Localization Method Using Graphical Summary of Multi-view RGB-D Images" The key contributions of the paper are: (i) Graphical summary generation from the multi-view RGB-D images of an indoor scene; (ii) Scene localization for an input query image through a graph summary matching approach.

The framework for MVGSL comprising of two major stages, namely Graphical Summary Generation, and Scene Localization.

The graph dataset https://drive.google.com/file/d/171YAnPZ1RESDE4o9kRyup_yLnT-xJ1eQ/view?usp=sharing is structured as follows:

buildingdata
└── building1
    ├── name               # building name
    ├── rooms         
    │   ├── name               # rooms name ex: kitchen, bedroom
    │   ├── rgbimages          # color images
    │   ├── depthimages        # depth images
    │   ├── salientobjects     # salient objects properties
    │       ├── labels     
    │       ├── Volume
    │       ├── centroid
    │       ├── corners    
    │       ├── orientation  
    │       └── objectPointClouds 
    │   ├── newadjacencyMatrix  %graph adjacency matrix
    │   ├── comp_adj  %complement of an adjacency matrix
    │   └── comp_graph     # graph consists of salient objects as nodes and comp_adj
└── ...

You can download the SUNRGBD data from https://rgbd.cs.princeton.edu/

SVG_cons: single-view graph construction
MVG_cons: multi-view graph construction
Matching: Scene localization
calculateEdfm: Comput Essential matrix
results folder contains the correspondence point to calculate E matrix.
The output folder contains the images of the generated SVG and MVG.
node2vec folder contains the embedding vectors for GT and query.

If you use the source code, please cite the following paper

@inproceedings{,
  title={AN INDOOR SCENE LOCALIZATION METHOD USING GRAPHICAL SUMMARY OF MULTI-VIEW RGB-D IMAGES},
  author={Meena, Preeti and Kumar, Himanshu and Yadav, Sandeep},
  booktitle={ICIP},
  pages={},
  year={2024},
  organization={IEEE}
}

• Sun RGB-D
• DFM
• node2vec