Home

Awesome

GPS-Net

LICENSE Python PyTorch

This repository contains data and code for the paper GPS-Net: Graph Property Sensing Network for Scene Graph Generation (CVPR 2020).

<div style="color:#0000FF" align="center"> <img src="images/framework.jpg"/> </div>

If the paper significantly inspires you, we request that you cite our work:

Bibtex

@inproceedings{lin2020gps,
  title={GPS-Net: Graph Property Sensing Network for Scene Graph Generation},
  author={Lin, Xin and Ding, Changxing and Zeng, Jinquan and Tao, Dacheng},
  booktitle={Proc. IEEE Conf. Comput. Vis. Pattern Recognit.},
  pages={3746--3753},
  year={2020}
}

Our reimplementations are based on rowanz/neural-motifs, and if you get any problem that cause you unable to run the project, you can check the issues under rowanz/neural-motifs first.

The Completed code is coming soon. But we have released some codes of DMP Modules, NPS-Loss and Frequency Softening. And you can see the codes in lib. Also, you can find more details in the draft version or our another implementation with a better object detection backbone.

Demo Video

Scene Graph Generation for Cityscape Video

Scene Graph Generation for Fruit Video

Scene Graph Generation for CCTV Video

Reported Results

R@20R@50R@100
SGGen22.628.431.7
SGCls36.139.240.1
PredCls60.766.968.8