sammica / changesim

A multi-modal, photo-realistic dataset for online end-to-end scene change detection and more (accepted to IROS2021).

ChangeSim

This repository provides ChangeSim dataset, codes and files for evaluation. Please refer to our paper (accepted to IROS2021) for more information about the dataset.

Recent updates

• Dataset download links
• Documentation for the dataset
• A tutorial for the visualization of ChangeSim
• A tutorial for the data collection using Airsim and UE4

Dataset download

The data is divided into train/test set and reference/query.

Reference_Sequence_Train(52.8 GB)

Reference_Sequence_Test(30.2 GB)

Query_Sequence_Train(42.8 GB)

Query_Sequence_Test(30.3 GB)

Data directory structure

Ref_Seq_
|
--- Warehouse_0                              # Environment folder
|       |
|       ---- Seq_0                           # Sequece
|       |      |
|       |      +--- rgb                      # 0.png - xxxx.png      
|       |      +--- depth                    # 0.png - xxxx.png
|       |      +--- semantic_segmentation    # 0.png - xxxx.png     
|       |      ---- raw                   
|       |      |     |
|       |      |     +--- rgb                # 0.png - xxxx.png
|       |      |     +--- depth              # 0.png - xxxx.png
|       |      |     ---- poses.g2o 
|       |      |     ---- rtabmap.yaml
|       |
|       +--- Seq_1
|
+-- Warehouse_1
.
.
+-- Warehouse_N



Query_Seq_
|
--- Warehouse_0                              # Environment folder
|       |
|       ---- Seq_0                           # Sequece
|       |      |
|       |      +--- rgb                      # 0.png - xxxx.png      
|       |      +--- depth                    # 0.png - xxxx.png
|       |      +--- semantic_segmentation    # 0.png - xxxx.png
|       |      +--- change_segmentation      # 0.png - xxxx.png
|       |      +--- pose                     # 0.txt - xxxx.txt
|       |      ---- t0                   
|       |      |     |
|       |      |     +--- rgb                # 0.png - xxxx.png
|       |      |     +--- depth              # 0.png - xxxx.png
|       |      |     +--- idx                # 0.txt - xxxx.txt
|       |      ---- cloud_map.ply
|       |      ---- trajectory.txt
|       |
|       +--- Seq_0_dust
|       .
|       .
|       +--- Seq_1_dark
|
+-- Warehouse_1
.
.
+-- Warehouse_N

Citation

If you find this project helpful, please consider citing this project in your publications. The following is the BibTeX of our work.

@inproceedings{park2021changesim,
author = {Park, Jin-Man and Jang, Jae-hyuk, and Yoo, Sahng-Min and Lee, Sun-Kyung and Kim, Ue-hwan and Kim, Jong-Hwan},
title = {ChangeSim: Towards End-to-End Online Scene Change Detection in Industrial Indoor Environments},
booktitle={2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2021},
organization = {IEEE},
url = {https://arxiv.org/abs/2103.05368},
}

Acknowledgement

This work was supported by Institute for Information & communications Technology Promotion (IITP) grant funded by the Korea government (MSIT) (No.2020-0-00440, Development of artificial intelligence technology that continuously improves itself as the situation changes in the real world).

changesim's Issues

unreal indoor environment

how can I get your unreal indoor environment?

Hello, when will all the codes be updated?

please help Where are the paired images ?

Hi
First of all thanks for the very useful dataset. I was wondering where are the paired images (which should be there with query images) in the query_test folder. I went to the t0 folder in one of the warehouse folder but found no files:

Can you please help?
Also one more question how can i get the pose information for each frame in reference seq?

Ref Sequence Poses

I found the poses defined in Trajectory.txt in Ref Sequence do not match images in either rgb or raw folder.
May I ask you if you have the correct poses for Ref Sequences?

RGBD frames

Hello,

Do you plan to, or could you provide raw images for the Query Sequences ? I noticed that the depth values were mapped from 0 to 255 and you didn't include the real distance values (in mm) for the Query Sequence frames.
Indeed, it is hard to make 3D reconstructions for the Query Sequences with mapped distance values.

Thanking you in advance for your response.