Publication Date

2025

Document Type

Dataset

Publisher

Edith Cowan University

School or Research Centre

School of Science

Description

The Panoramic Scene Change Understanding (PanoSCU) dataset supports eight research tasks, encompassing both image-level and scene-level panorama settings. These tasks include local and panoramic object detection, segmentation and change understanding, object tracking, and change reversal. Currently available 2D change understanding datasets primarily concentrate on two separate tasks: change detection (changes localization) and change captioning (change content description). Our PanoSCU, targets the simultaneous localization and description of object changes in indoor panoramic 2D images. Although the PanoSCU dataset is artificially generated using the Ai2Thor simulator, it is highly complex. It incorporates 2650 rearrangement scenarios (panorama pairs) with 48 indoor object classes of different sizes. Since the panoramas in PanoSCU are generated from different amounts of images for each scene, they significantly vary in ratios and sizes. This variability presents a challenge for current algorithms, which typically operate with standard-sized image inputs. Moreover, the source images for panoramas are captured by an agent at different times of the day, with different light sources, and from slightly different locations, further increasing the dataset's complexity.

DOI

10.25958/4p6j-z657

Methodology

The data was generated using Ai2Thor simulator.

Start of data collection time period

2023

End of data collection time period

2025

File Format(s)

txt, png, json

File Size

154 GB

Creative Commons License

Creative Commons Attribution-Noncommercial 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License.

Contact

mariia.khan@ecu.edu.au

Download
train.pkl.gz (2175 kB)
test.pkl.gz (564 kB)
val.pkl.gz (512 kB)
FloorPlan26__test__1.txt (1 kB)

Share

 
COinS