Taehyeon Kim and Byung-Cheol Min.

Project Page

Code repository for Semantic Layering in Room Segmentation via LLMs (SeLRoS). This repository includes 2D Map generation code and Room Information Interpreter code, and a data set containing ground truth, object information file, top view image, and room segmentation results for each environment for an experiment in 30 environments.

Install dependencies:

pip install -r requirments.txt

For more details, please check the following link ProcTHOR.

Run the following command to interpret segmenation map and generate output.

python3 room_info_interpreter.py --input1 {segmented_map.png} --input2 {related_information.txt}

Note: You can use the segmented_map.png and related_information.txt in data\vrf\.

Run the following script to generate top view image and 2d map of environment.

python3 run_get_2d_map.py

Run the following script to get scenes of specific position.

python3 run_get_scenes.py

Note: If you want to change observation position, change x, z value in 52 line.

The repository contains ground truth, object information file, top view image, and room segmentation results for each environment.

Refer to data\ground_truth\ for the ground truth of each environment.

Refer to data\object_information\ for the object information txt file of each room in each environment.

Refer to data\selros\ for the final room segmentation results using SeLRoS.

Refer to data\top_view\ for the top view image of each environment.

Refer to data\vrf\ for the room segmentation results using Voronoi Random Field algorithm and related information (each segmented room's center coordinate and color).

If you find this work useful for your research, please consider citing:

@article{kim2024semantic,
  title={Semantic Layering in Room Segmentation via LLMs},
  author={Kim, Taehyeon and Min, Byung-Cheol},
  journal={arXiv preprint arXiv:2403.12920},
  year={2024}
}