T-STAR: A Large-Scale Benchmark for Spatio-Temporal Panoptic Scene Graph Generation in Satellite Video
Given a satellite video sequence, TPSG aims to generate a structured graph composed of a set of triplets ⟨subject, relationship, object⟩ with explicit temporal spans, thereby describing dynamic geospatial scenes by jointly modeling identity-consistent instance masks and spatio-temporal relationships among panoptic scene elements.
TPSGdemo15.mp4
Illustration of spatio-temporal panoptic scene graph generation (TPSG) in satellite video.
We construct T-STAR, the first large-scale dataset for spatio-temporal panoptic scene graph generation (TPSG) in satellite video. It contains more than 1.1 million instance masks and over 3.8 million spatio-temporal triplets across 39 fine-grained object categories and 70 fine-grained relationship categories.
Scenes
Representative annotated scenes from the T-STAR dataset.
Object
Statistics and visualization of objects from the T-STAR dataset.
Relationship
Statistics and visualization of relationships from the T-STAR dataset.
Obj-Rel Interaction
Interaction mapping among subject categories, relationship categories, and object categories.
Download Links
- T-STAR Dataset: Coming soon.
If you find this work helpful for your research, please consider giving this repo a star ⭐ and citing our papers:
# SGG in Large-Size Satellite Imagery
@ARTICLE{STAR,
author={Li, Yansheng and Wang, Linlin and Wang, Tingzhu and Yang, Xue and Luo, Junwei and Wang, Qi and Deng, Youming and Wang, Wenbin and Sun, Xian and Li, Haifeng and Dang, Bo and Zhang, Yongjun and Yu, Yi and Yan, Junchi},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={STAR: A First-Ever Dataset and a Large-Scale Benchmark for Scene Graph Generation in Large-Size Satellite Imagery},
year={2025},
volume={47},
number={3},
pages={1832-1849},
keywords={Stars;Annotations;Marine vehicles;Satellite images;Visualization;Object detection;Cognition;Benchmark testing;Complexity theory;Bridges;Large-size satellite imagery;object detection;relationship prediction;scene graph generation benchmark},
doi={10.1109/TPAMI.2024.3508072}}
Our code is based on OpenPVSG in CVPR'23, we sincerely thank them.